Merge pull request 'issue 215: Use OpenAI official API to list available models' (#58) from bugfix/fix-list-models into master

Reviewed-on: https://192.168.8.63/git/darthrubik/drharness/pulls/58

CMakeLists.txt

@@ -32,6 +32,8 @@ add_library(drharness_lib
     src/debug_client.cpp
     src/editor.cpp
     src/skill.cpp
+   src/rebase_engine.cpp
+    src/utility.cpp
     src/base64_util.cpp
     ${TOOL_SOURCES}
 )

CONTRIBUTING.md

@@ -94,10 +94,19 @@ Exception only if it is API calling code which is hard to test
 deterministically.
 
 ### Adding new flags to the CLI for testing
-Adding new flags to the CLI for testing is allowed and ENCOURAGED!  For example
-for the skill stuff, we added a flag -D that just dumps all skill directories
-found.  This is useless to the end user, but lets us test the fact that our
-skill directory search was working.
+Adding new flags to the CLI for testing is allowed and ENCOURAGED!  These
+flags serve as **hooks into internal functionality** to enable testing of
+modules that don't have a natural CLI entry point.
+
+For example, for the skill stuff, we added a flag -D that just dumps all skill
+directories found.  This is useless to the end user, but lets us test the fact
+that our skill directory search was working.
+
+**Test flags should be minimal hooks:**
+- They should accept input, call the internal function, and output the result
+- They should NOT contain test logic (assertions, validation, etc.)
+- All test logic belongs in the Bash test scripts in `tests/`
+- The flag's only job is to bridge CLI arguments to the internal function
 
 If you add a new flag for testing you don't need to update the README.md (but
 updating the -h message is still mandatory).

README.md

@@ -1,6 +1,6 @@
 # drharness
 
-A C++20 LLM (Open AI Compatible) harness designed for agentic workflows, file
+A C++23 LLM (Open AI Compatible) harness designed for agentic workflows, file
 manipulation, and task delegation. It implements a tool-use loop that allows a
 large language model to dynamically invoke shell commands, edit files, and
 delegate sub-tasks to other specialized agents. (I wouldn't use the agents
@@ -17,11 +17,11 @@ to it)
 *   **`Client Layer`**: 
     *   `OllamaClient`: Handles the actual communication with an Ollama-compatible API server via HTTP (using `libcurl`).
     *   `DebugClient`: A CLI-friendly mock client that reads a mock JSON response from `stdin`, allowing for deterministic testing of harness logic without relying on an external LLM.
-*   **State Management**: Conversations are serialized to and deserialized from JSON files. This includes the full message history (`history`), the active tool registry (`tools`), and harness configuration (`config`).
+*   **State Management**: Conversations are serialized to and deserialized from JSON files. This includes the full message history (`history`), the active tool registry (`tools`), and harness configuration (`config` including `base_url`, `model`, `max_tool_iterations`, `verbose`, and `depth`).
 
 ## 🛠️ How to Build
 
-The project requires CMake version 3.16 or newer, C++20 support, `curl`, and the `nlohmann_json` library.
+The project requires CMake version 3.16 or newer, C++23 support, `curl`, and the `nlohmann_json` library.
 
 ```bash
 # Configure the build
@@ -75,6 +75,7 @@ configuration.
 *   `-s <name>`: Manually constrain the harness to a specific subset of skills (can be specified multiple times)
 *   `-i <path>`: Image file to attach to the prompt (can be specified multiple times)
 *   `-m <prompt>`: The user prompt to send to the agent.
+*   `-P <file>`: Read the prompt from a file instead of the command line.
 *   `-M <model>`: Override the Ollama model (e.g., `mistral`, `llama3`).
 
 ## ✍️ Editor Integration
@@ -96,6 +97,9 @@ export EDITOR=nvim  # or vscode, nano, vim, emacs, etc.
 
 *Note: If neither `VISUAL` nor `EDITOR` is set, the harness will refuse to launch the editor and throw an error instead.*
 
+
+For detailed documentation on history editing, see [docs/history-editing.md](docs/history-editing.md).
+
 ## 🐳 Docker Isolation
 
 The harness can be compiled and run inside an isolated Alpine Linux container.

docs/history-editing.md

@@ -0,0 +1,83 @@
+## 📜 History Editing
+
+If a conversation goes off track, you can surgically edit your conversation history using the interactive history editing mode. This allows you to rewind to a specific point, modify messages, or remove problematic exchanges without starting over.
+
+### Invoking History Editing
+
+Use the `-b` or `--edit-history` flag along with `-F` to specify your chat file:
+
+```bash
+./build/drharness -F conversation.json -b
+```
+
+*Note: The `-F` flag is required because editing requires an existing chat file.*
+
+### The Rebase Editor Format
+
+When invoked, the harness opens your configured editor with all messages in your conversation listed in order. Each line follows a git rebase-style format:
+
+```
+<command> <index> [<role>] <display_content>
+```
+
+**Commands:**
+- `pick`: Keep the message unchanged
+- `edit`: Modify the message content and/or role
+- `drop`: Delete the message (or simply remove the line)
+
+**Role:** (optional, only for `edit` commands)
+- `[user]`
+- `[assistant]`
+- `[tool]`
+
+**Display content:** The first line of the message (for reference only)
+
+Lines starting with `#` are comments and are ignored. Empty lines are treated as drops.
+
+### Editing Message Content and Roles
+
+When you mark a message with `edit`, the harness will open your editor again to let you modify the content. The edit prompt shows:
+- The previous message as context (for reference)
+- The `>8` separator
+- The message content you can edit
+
+You can change the role by modifying the role in the rebase line (e.g., `edit 2 [user]`).
+
+### Common Workflows
+
+**Editing a single message:**
+```
+pick 0 [user] Hello, can you help me?
+pick 1 [assistant] Sure! What do you need?
+edit 2 [assistant] I misunderstood. Let me try again.
+pick 3 [user] That's better, thanks!
+```
+Then, when editing message 2, you'll be prompted to enter the new content.
+
+**Dropping unwanted messages:**
+```
+pick 0 [user] Start the project
+drop 1 [assistant] Wrong approach
+pick 2 [user] Let me try a different way
+```
+Or simply delete the line:
+```
+pick 0 [user] Start the project
+# pick 1 [assistant] Wrong approach
+pick 2 [user] Let me try a different way
+```
+
+**Modifying message roles:**
+```
+pick 0 [user] Write a function
+edit 1 [user] I meant write a class instead
+```
+
+**Continuing a conversation after editing:**
+After saving and exiting the editor, your changes are persisted to the chat file. The program exits, allowing you to inspect the result. To continue the conversation:
+```bash
+./build/drharness -F conversation.json -m "Continue from where we left off"
+```
+
+*Note: The history editor uses strict validation. Any syntax error (invalid command, out-of-range index, invalid role) will cause the editor to abort with an aggregated error message. No partial changes are applied.*
+

include/json_types.hpp

@@ -362,8 +362,10 @@ struct HarnessConfig {
     size_t max_tool_iterations = 16;
     bool   verbose             = false;
     size_t depth               = 0;
+    std::string base_url       = "https://192.168.8.63/ollama";
+    std::string model          = "qwen3.6:35b-a3b-128k";
 };
-DRH_DEFINE_TYPE_OPTIONAL(HarnessConfig, max_tool_iterations, verbose, depth)
+DRH_DEFINE_TYPE_OPTIONAL(HarnessConfig, max_tool_iterations, verbose, depth, base_url, model)
 
 struct FunctionCall {
     std::string name = "";
@@ -503,14 +505,14 @@ struct ChatResponse {
 DRH_DEFINE_TYPE_MIXED(ChatResponse, (choices), (usage))
 
 struct ModelInfo {
-    std::string name;
+    std::string id;
 };
-DRH_DEFINE_TYPE_REQUIRED(ModelInfo, name)
+DRH_DEFINE_TYPE_REQUIRED(ModelInfo, id)
 
 struct ModelListResponse {
-    std::vector<ModelInfo> models;
+    std::vector<ModelInfo> data;
 };
-DRH_DEFINE_TYPE_REQUIRED(ModelListResponse, models)
+DRH_DEFINE_TYPE_REQUIRED(ModelListResponse, data)
 
 // Tool parameter structs
 
@@ -548,8 +550,12 @@ struct ShellToolParams {
 DRH_DEFINE_TYPE_OPTIONAL(ShellToolParams, command, timeout_seconds)
 
 // Safe JSON parsing - returns discarded JSON on parse failure
-inline nlohmann::json parse_json_safely(const std::string& input) {
-    return nlohmann::json::parse(input, nullptr, false);
+inline std::expected<nlohmann::json, std::string> parse_json_safely(const std::string& input) {
+    auto j = nlohmann::json::parse(input, nullptr, false);
+    if (j.is_discarded()) {
+        return std::unexpected("error: invalid JSON");
+    }
+    return j;
 }
 
 // Safe regex construction - returns error on invalid pattern

include/ollama_client.hpp

@@ -9,15 +9,13 @@
 using http_req_cb_t = std::function<std::string(std::string const& old, std::string const& new_str)>;
 
 struct OllamaConfig {
-    std::string base_url = "https://192.168.8.63/ollama";
-    std::string model    = "qwen3.6:35b-a3b-128k";
     std::string auth_token = "";
     int         timeout_s = 60*10;
 };
 
 class OllamaClient : public Client {
 public:
-    explicit OllamaClient(OllamaConfig config = {});
+    explicit OllamaClient(std::string base_url, std::string model, OllamaConfig config = {});
     ~OllamaClient() override;
 
     std::expected<Message, std::string> chat(const std::vector<Message>& messages,
@@ -26,6 +24,8 @@ public:
     std::expected<std::vector<std::string>, std::string> list_models() const override;
 
 private:
+    std::string base_url_;
+    std::string model_;
     OllamaConfig config_;
     nlohmann::json build_body(const std::vector<Message>& messages,
                               const std::vector<Tool>& tools) const;

include/rebase_engine.hpp

@@ -0,0 +1,83 @@
+#pragma once
+#include "json_types.hpp"
+#include <string>
+#include <vector>
+#include <expected>
+
+// Forward declaration
+struct Message;
+
+enum class RebaseCommand { Pick, Edit, Drop };
+
+// JSON serialization for RebaseCommand
+inline void to_json(nlohmann::json& j, RebaseCommand cmd) {
+    switch (cmd) {
+        case RebaseCommand::Pick: j = "pick"; break;
+        case RebaseCommand::Edit: j = "edit"; break;
+        case RebaseCommand::Drop: j = "drop"; break;
+    }
+}
+
+template<>
+inline std::expected<RebaseCommand, std::string> json_to<RebaseCommand>(const nlohmann::json& j) {
+    if (!j.is_string()) return std::unexpected("expected string for RebaseCommand");
+    std::string s = j.get<std::string>();
+    if (s == "pick") return RebaseCommand::Pick;
+    if (s == "edit") return RebaseCommand::Edit;
+    if (s == "drop") return RebaseCommand::Drop;
+    return std::unexpected("invalid RebaseCommand: '" + s + "'");
+}
+
+template<> struct has_json_to<RebaseCommand> : std::true_type {};
+
+struct RebaseLine {
+    RebaseCommand command;
+    size_t index;
+    std::string role;
+    std::string content;  // For edit commands, stores the new content
+    bool content_edited = false;  // True if content was explicitly edited (even if empty)
+};
+
+DRH_DEFINE_TYPE_MIXED(RebaseLine, (command, index, role), (content))
+
+class RebaseEngine {
+public:
+    std::expected<std::vector<RebaseLine>, std::string> parse_rebase_file(
+        const std::string& content, size_t history_size) const;
+    
+    std::expected<std::vector<Message>, std::string> apply_rebase(
+        const std::vector<RebaseLine>& lines,
+        const std::vector<Message>& history) const;
+};
+
+class HistoryEditor {
+public:
+    // Build the rebase editor content from a message history
+    // Each message is formatted as: pick N [role] first_line_of_content
+    // Tool calls display as: [tool_call: function_name]
+    // Empty messages display as: (empty)
+    std::string build_rebase_content(const std::vector<Message>& history) const;
+
+    // Build the edit prompt for editing a single message
+    // Shows previous message as context, separator, and current content to edit
+    // For message 0, shows "(no previous message)" instead of previous message
+    std::string build_edit_prompt(const std::vector<Message>& history,
+                                   size_t edit_index,
+                                   const std::string& role) const;
+
+    // Full interactive edit flow
+    // 1. Build rebase content and spawn editor
+    // 2. Parse and validate rebase commands
+    // 3. For each edit command, build edit prompt and spawn editor
+    // 4. Apply all changes
+    // Returns modified history on success, error message on failure
+    std::expected<std::vector<Message>, std::string> edit_history_interactive(
+        const std::vector<Message>& history) const;
+
+private:
+    // Extract the first line of content from a message
+    std::string get_first_line(const std::string& content) const;
+
+    // Format display text for a single message
+    std::string get_message_display(const Message& msg) const;
+};

include/utility.hpp

@@ -0,0 +1,31 @@
+#pragma once
+#include <string>
+#include <expected>
+#include <functional>
+#include <atomic>
+
+extern std::atomic<bool> ctrlc_request;
+
+// Trim whitespace from both ends of a string
+std::string trim(const std::string& s);
+
+// Generate a random alphanumeric string
+std::string random_string(size_t length = 8);
+
+// Generate a random temp file path
+std::string random_file_name(size_t length = 8);
+
+// RAII defer helper class for cleanup
+class Defer {
+    std::function<void()> fn;
+
+public:
+    Defer(std::function<void()> f) : fn(f) {}
+    ~Defer() { fn(); }
+};
+
+// Spawn an editor with initial content and return the content after the separator (-- >8 --)
+std::expected<std::string, std::string> spawn_editor(const std::string& initial_content);
+
+// Parse an unsigned long from a C-string, returning an error if parsing fails
+std::expected<unsigned long, std::string> string_toul(std::string const& str);

src/debug_client.cpp

@@ -12,10 +12,11 @@ std::expected<Message, std::string> DebugClient::chat(const std::vector<Message>
         return std::unexpected("DebugClient failed to read from stdin");
     }
     
-    auto j = parse_json_safely(input);
-    if (j.is_discarded()) {
-        return std::unexpected("DebugClient invalid JSON input: parse error");
+    auto j_result = parse_json_safely(input);
+    if (!j_result) {
+        return std::unexpected("DebugClient invalid JSON input: " + j_result.error());
     }
+    auto j = j_result.value();
     auto result = json_to<ChatResponse>(j);
     if (!result) {
         return std::unexpected(std::string("DebugClient invalid JSON input: ") + result.error());

src/editor.cpp

@@ -1,138 +1,38 @@
 #include "editor.hpp"
 #include "ollama_client.hpp"
-#include <fstream>
-#include <iostream>
+#include "utility.hpp"
 #include <sstream>
-#include <cstdlib>
 #include <string>
-#include <algorithm>
-#include <filesystem>
-#include <random>
-#include <functional>
-
-static std::string trim(const std::string& s) {
-    auto start = s.find_first_not_of(" \t\n\r");
-    return (start == std::string::npos) ? "" : s.substr(start);
-}
-
-static std::string random_string(size_t length = 8) {
-    static const char chars[] =
-        "abcdefghijklmnopqrstuvwxyz"
-        "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
-        "0123456789";
-
-    static std::random_device rd;
-    static std::mt19937 gen(rd());
-    static std::uniform_int_distribution<> dis(0, sizeof(chars) - 2);
-
-    std::string s;
-    for (size_t i = 0; i < length; ++i) {
-        s += chars[dis(gen)];
-    }
-    return s;
-}
-
-static std::string random_file_name(size_t length = 8)
-{
-    namespace fs = std::filesystem;
-
-    fs::path temp_dir = fs::temp_directory_path();
-    fs::path temp_file = temp_dir / ("tmp_" + random_string(length));
-
-
-    return temp_file.string();
-}
-
-class Defer {
-    std::function<void()> fn;
-
-public:
-    Defer(std::function<void()> f) : fn(f) {}
-    ~Defer() { fn(); }
-};
-
 
 std::expected<std::string, std::string> run_editor(HarnessState state, bool full_history) {
-    auto file_path = random_file_name();
-    auto cleanup = Defer([&](){
-        std::filesystem::remove(file_path);
-    });
+    std::ostringstream initial_content;
 
-    {
-        std::ofstream file(file_path);
-        if (!file.is_open()) {
-            return std::unexpected("Failed to open temp file for editor.");
-        }
-
-        if (state.history.empty()) {
-            file << "New Conversation.\n\n";
-            file << "This is the start of a new chat. Type your prompt below the separator.\n";
-            file << "The AI will respond after you save and close the editor.\n\n";
+    if (state.history.empty()) {
+        initial_content << "New Conversation.\n\n";
+        initial_content << "This is the start of a new chat. Type your prompt below the separator.\n";
+        initial_content << "The AI will respond after you save and close the editor.\n\n";
+    } else {
+        nlohmann::json j = nlohmann::json(state);
+        if (full_history) {
+            std::ostringstream history_str;
+            print_history(history_str, j, false);
+            initial_content << history_str.str();
         } else {
-            nlohmann::json j = nlohmann::json(state);
-            if (full_history) {
-                print_history(file, j, false);
-            } else {
-                // Find the last AI response
-                const Message* last_ai_msg = nullptr;
-                for (int i = static_cast<int>(state.history.size()) - 1; i >= 0; --i) {
-                    if (state.history[i].role == "assistant") {
-                        last_ai_msg = &state.history[i];
-                        break;
-                    }
-                }
-                
-                if (last_ai_msg) {
-                    file << "[" << last_ai_msg->role << "] ";
-                    if (!last_ai_msg->reasoning.empty()) {
-                        file << "[Thinking: " << last_ai_msg->reasoning << "] ";
-                    }
-                    file << last_ai_msg->content << "\n\n";
-                } else {
-                    // Fallback to last message if no assistant message found
-                    const auto& last_msg = state.history.back();
-                    file << "[" << last_msg.role << "] " << last_msg.content << "\n\n";
-                }
-            }
+            const auto& last_msg = state.history.back();
+            initial_content << "[" << last_msg.role << "] " << last_msg.content << "\n\n";
         }
-
-        file << "-- >8 --\n\n";
-        file.close();
-
-        const char* editor_env = getenv("VISUAL");
-        if (!editor_env) editor_env = getenv("EDITOR");
-        if (!editor_env) {
-            return std::unexpected("Neither $VISUAL nor $EDITOR is set.");
-        }
-
-        std::string cmd = std::string(editor_env) + " " + file_path;
-        int res = system(cmd.c_str());
-        if (res != 0) {
-            return std::unexpected("Editor exited with error.");
-        }
-
-        std::ifstream infile(file_path);
-        std::string content((std::istreambuf_iterator<char>(infile)), std::istreambuf_iterator<char>());
-        
-        const std::string seperator = "\n-- >8 --\n";
-        size_t pos = content.find(seperator);
-        if (pos == std::string::npos) {
-            return std::unexpected("Scissors not found.");
-        }
-        
-        size_t last_pos = std::string::npos;
-        while (pos != std::string::npos) {
-            last_pos = pos;
-            pos = content.find(seperator, pos+1);
-        }
-        
-        std::string extracted = content.substr(last_pos + seperator.size());
-        extracted = trim(extracted);
-        
-        if (extracted.empty()) {
-            return std::unexpected("Prompt is empty after separator.");
-        }
-        
-        return extracted;
     }
+
+    initial_content << "-- >8 --\n\n";
+
+    auto result = spawn_editor(initial_content.str());
+    if (!result) {
+        return result;
+    }
+
+    if (result->empty()) {
+        return std::unexpected("Prompt is empty after separator.");
+    }
+
+    return *result;
 }

src/harness.cpp

@@ -2,6 +2,7 @@
 #include "json_types.hpp"
 #include "base64_util.hpp"
 #include <iostream>
+#include "utility.hpp"
 
 Harness::Harness(std::unique_ptr<Client> client, const ToolRegistry& registry,
                  HarnessConfig config)
@@ -80,11 +81,31 @@ std::expected<std::string, std::string> Harness::chat(const Message& user_messag
     push_message(user_message);
 
     for (size_t i = 0; i < config_.max_tool_iterations; ++i) {
+
+        if (ctrlc_request)
+        {
+            // Cancellation (Ctrl+C): return empty string to indicate the chat was
+            // interrupted. The user message has already been pushed to history, so
+            // the conversation state will be saved to the chat file. This allows
+            // the user to resume the conversation from this point.
+            //
+            // Note: We return an empty string (success) rather than std::unexpected
+            // because cancellation is not an error condition - it's a deliberate
+            // user action to stop processing. Using std::unexpected would prevent
+            // the harness state from being saved, losing the user's input.
+            return "";
+        }
+
         Message response;
         bool done = false;
         size_t retries = 20;
         while (!done)
         {
+            if (ctrlc_request)
+                // Cancellation during API call: same semantics as above - return
+                // empty string to preserve conversation state in the chat file.
+                return "";
+
             retries--;
             if (retries == 0)
             {
@@ -114,12 +135,12 @@ std::expected<std::string, std::string> Harness::chat(const Message& user_messag
                 if (config_.verbose)
                     std::cerr << prefix << "[tool call] " << tool_name << "(" << args_str << ")\n";
 
-                Content result = "";
-                auto args_json = parse_json_safely(args_str);
-                if (args_json.is_discarded()) {
+               Content result = "";
+                auto args_json_result = parse_json_safely(args_str);
+                if (!args_json_result) {
                     result = "Error: invalid JSON in tool arguments";
                 } else {
-                    result = registry_.invoke(tool_name, args_json);
+                    result = registry_.invoke(tool_name, args_json_result.value());
                 }
 
                 if (config_.verbose)
@@ -140,6 +161,12 @@ std::expected<std::string, std::string> Harness::chat(const Message& user_messag
                 tool_result.content = result;
                 push_message(tool_result);
             }
+
+            if (ctrlc_request)
+                // Cancellation after tool execution: same semantics as above -
+                // return empty string to preserve conversation state in the chat file.
+                return "";
+
             continue;
         }
 

src/main.cpp

@@ -5,6 +5,8 @@
 #include "editor.hpp"
 #include "tool_registry.hpp"
 #include "agent_profile.hpp"
+#include "rebase_engine.hpp"
+#include "utility.hpp"
 #include "base64_util.hpp"
 #if !defined(_WIN32)
 #include "shell_tool.hpp"
@@ -27,6 +29,47 @@
 #include <filesystem>
 #include <random>
 #include <iomanip>
+#ifndef _WIN32
+#include <signal.h>
+#else
+#include <windows.h>
+#endif
+
+std::atomic<bool> ctrlc_request;
+
+
+void cancel_request(void)
+{
+    if (ctrlc_request)
+    {
+        exit(1); // Already asked to quit once, lets immediately die
+    }
+    ctrlc_request = true;
+}
+
+#if defined(_WIN32)
+BOOL WINAPI console_handler(DWORD ctrl_type) {
+    if (ctrl_type == CTRL_C_EVENT) {
+        cancel_request();
+        return TRUE;
+    }
+    return FALSE;
+}
+
+void setup_signal_handler() {
+    SetConsoleCtrlHandler(console_handler, TRUE);
+}
+#else
+void signalHandler(int signal) {
+    if (signal == SIGINT) {
+        cancel_request();
+    }
+}
+
+void setup_signal_handler() {
+    signal(SIGINT, signalHandler);
+}
+#endif
 
 static const std::vector<AgentProfile> profiles = {
 #if !defined(_WIN32)
@@ -76,10 +119,34 @@ static std::string generate_uuid() {
     return ss.str();
 }
 
+// Helper function to read and deserialize HarnessState from a file
+// Returns the parsed HarnessState, or exits on error
+static HarnessState read_harness_state_or_exit(const char* file_path) {
+    std::ifstream file(file_path);
+    if (!file.is_open()) {
+        std::cerr << "Error: Cannot open file " << file_path << std::endl;
+        exit(1);
+    }
+    std::string file_content((std::istreambuf_iterator<char>(file)), std::istreambuf_iterator<char>());
+    auto data = expect_or_exit(parse_json_safely(file_content));
+    auto state_result = json_to<HarnessState>(data);
+    if (!state_result) {
+        std::cerr << "Error: Failed to parse chat file: " << state_result.error() << std::endl;
+        exit(1);
+    }
+    return state_result.value();
+}
+
 int main(int argc, char** argv) {
+    setup_signal_handler();
+
+    HarnessConfig harnessConfig = HarnessConfig{
+        .max_tool_iterations = 1000,
+        .verbose             = true,
+    };
     OllamaConfig cfg;
-    if (const char* m = std::getenv("OLLAMA_MODEL")) cfg.model    = m;
-    if (const char* u = std::getenv("OLLAMA_URL"))   cfg.base_url = u;
+    if (const char* m = std::getenv("OLLAMA_MODEL")) harnessConfig.model    = m;
+    if (const char* u = std::getenv("OLLAMA_URL"))   harnessConfig.base_url = u;
     if (const char* a = std::getenv("OLLAMA_AUTHORIZATION_TOKEN")) cfg.auth_token = a;
 
     bool use_debug = false;
@@ -91,6 +158,14 @@ int main(int argc, char** argv) {
     char const* test_expect_exit_error = nullptr;
     char const* test_type_required_type = nullptr;
     char const* test_type_required_json = nullptr;
+   char const* test_rebase_history_size = nullptr;
+    char const* test_rebase_content = nullptr;
+    char const* test_build_rebase_json = nullptr;
+    char const* test_apply_rebase_lines = nullptr;
+    char const* test_apply_rebase_history = nullptr;
+    char const* test_edit_prompt_json = nullptr;
+    char const* test_edit_prompt_index = nullptr;
+    char const* test_edit_prompt_role = nullptr;
     char const* test_base64_encode = nullptr;
     char const* test_base64_decode = nullptr;
     char const* test_mime_type = nullptr;
@@ -112,7 +187,7 @@ int main(int argc, char** argv) {
     // Add more base tools here:
 
     for (const auto& profile : profiles)
-        expect_or_exit(master_registery.register_tool(make_agent_tool(profile, master_registery, cfg.base_url)));
+        expect_or_exit(master_registery.register_tool(make_agent_tool(profile, master_registery, harnessConfig.base_url)));
 
     auto consume_arg = [&](){
         argc--;
@@ -129,11 +204,13 @@ int main(int argc, char** argv) {
     char const* file_to_open = nullptr;
     std::string auto_temp_file;  // Path to auto-generated temp file (if any)
     char const* prompt = nullptr;
+    char const* prompt_file = nullptr;
     bool start_new = false;
     bool list_models = false;
     char const* print_file = nullptr;
     std::vector<std::string> tools;
     bool full_history = false;
+    bool edit_history = false;
     std::vector<std::string> skill_filters;
     std::vector<std::string> image_paths;
 
@@ -166,10 +243,15 @@ int main(int argc, char** argv) {
             consume_arg();
             prompt = consume_arg();
         }
+        else if (argv[0] == std::string("-P"))
+        {
+            consume_arg();
+            prompt_file = consume_arg();
+        }
         else if (argv[0] == std::string("-M"))
         {
             consume_arg();
-            cfg.model = consume_arg();
+            harnessConfig.model = consume_arg();
         }
         else if (argv[0] == std::string("-d"))
         {
@@ -212,15 +294,39 @@ int main(int argc, char** argv) {
             test_type_required_type = consume_arg();
             test_type_required_json = consume_arg();
         }
+        else if (argv[0] == std::string("-z"))
+        {
+            consume_arg();
+            test_rebase_history_size = consume_arg();
+            test_rebase_content = consume_arg();
+        }
+        else if (argv[0] == std::string("-Z"))
+        {
+            consume_arg();
+            test_build_rebase_json = consume_arg();
+        }
+        else if (argv[0] == std::string("-A"))
+        {
+            consume_arg();
+            test_apply_rebase_lines = consume_arg();
+            test_apply_rebase_history = consume_arg();
+        }
         else if (argv[0] == std::string("-B"))
         {
             consume_arg();
             test_base64_encode = consume_arg();
         }
-        else if (argv[0] == std::string("-b"))
+        else if (argv[0] == std::string("-K"))
         {
             consume_arg();
             test_base64_decode = consume_arg();
+        }
+         else if (argv[0] == std::string("-W"))
+        {
+            consume_arg();
+            test_edit_prompt_json = consume_arg();
+            test_edit_prompt_index = consume_arg();
+            test_edit_prompt_role = consume_arg();
         }
         else if (argv[0] == std::string("-T"))
         {
@@ -256,6 +362,11 @@ int main(int argc, char** argv) {
             consume_arg();
             skill_filters.push_back(consume_arg());
         }
+     else if (argv[0] == std::string("-b") || argv[0] == std::string("--edit-history"))
+        {
+            consume_arg();
+            edit_history = true;
+        }
         else if (argv[0] == std::string("-i"))
         {
             consume_arg();
@@ -277,8 +388,11 @@ int main(int argc, char** argv) {
                       << "  -i <path>   Image file to attach (can be specified multiple times)\n"
                       << "  -n          Start a new conversation (auto-generates temp file if -F not specified)\n"
                       << "  -m <prompt> Prompt message (required)\n"
+                      << "  -P <file>   Read prompt from file\n"
                       << "  -M <model>  Ollama model name\n"
                       << "  -f          Open editor with full history (requires -F)\n"
+                      << "  -b          Edit history (requires -F)\n"
+                      << "  --edit-history\n"
                       << "  -l          List available tools\n"
                       << "  -L          List available models on server side\n"
                       << "  -D          Debug: print all discovered skill paths and exit\n"
@@ -288,8 +402,12 @@ int main(int argc, char** argv) {
                       << "  -E <string> Debug: test expect_or_exit success case and exit\n"
                       << "  -e <string> Debug: test expect_or_exit error case and exit\n"
                       << "  -R <type> <json> Debug: test json_to<Type>() for required fields and exit\n"
+                      << "  -z <size> <content> Debug: test rebase parser and exit\n"
+                      << "  -Z <json>   Debug: test build_rebase_content with JSON history and exit\n"
+                      << "  -A <lines> <history> Debug: test apply_rebase with JSON and exit\n"
+                      << "  -W <json> <index> <role> Debug: test build_edit_prompt and exit\n"
                       << "  -B <string> Debug: test base64 encode and exit\n"
-                      << "  -b <string> Debug: test base64 decode and exit\n"
+                      << "  -K <string> Debug: test base64 decode and exit\n"
                       << "  -T <ext>    Debug: test MIME type detection and exit\n"
                       << "  -U <ext> <b64> Debug: test data URI generation and exit\n"
                       << "  -I <file>   Debug: test image file reading and exit\n"
@@ -304,11 +422,49 @@ int main(int argc, char** argv) {
         }
     }
 
+    // Validate -b / --edit-history flag requirements
+    if (edit_history && !file_to_open)
+    {
+        std::cerr << "Error: -b/--edit-history requires -F <chat_file> to be specified" << std::endl;
+        return 1;
+    }
+
+    // Validate -b / --edit-history flag with empty history
+    if (edit_history && file_to_open)
+    {
+        HarnessState state_data = read_harness_state_or_exit(file_to_open);
+        
+        if (state_data.history.empty()) {
+            std::cerr << "Error: Chat file has no history, cannot edit" << std::endl;
+            return 1;
+        }
+        
+        // Run the interactive edit flow
+        HistoryEditor editor;
+        auto edit_result = editor.edit_history_interactive(state_data.history);
+        if (!edit_result) {
+            std::cerr << "Error: " << edit_result.error() << std::endl;
+            return 1;
+        }
+        
+        // Save modified history to chat file
+        state_data.history = edit_result.value();
+        std::ofstream outfile(file_to_open);
+        if (!outfile.is_open()) {
+            std::cerr << "Error: Cannot write to file " << file_to_open << std::endl;
+            return 1;
+        }
+        outfile << nlohmann::json(state_data);
+        
+        // Exit after successful edit
+        return 0;
+    }
+
     // Helper function to create the client based on the debug flag
     auto create_client = [&](bool debug) -> std::unique_ptr<Client> {
         if (debug)
             return std::make_unique<DebugClient>();
-        return std::make_unique<OllamaClient>(cfg);
+        return std::make_unique<OllamaClient>(harnessConfig.base_url, harnessConfig.model, cfg);
     };
 
     // Scan for skills
@@ -364,11 +520,7 @@ int main(int argc, char** argv) {
 
     if (test_json_parse) {
         auto result = parse_json_safely(test_json_parse);
-        if (result.is_discarded()) {
-            std::cout << "discarded" << std::endl;
-        } else {
-            std::cout << "valid" << std::endl;
-        }
+        std::cout << expect_or_exit(result).dump(2) << std::endl;
         return 0;
     }
 
@@ -386,23 +538,56 @@ int main(int argc, char** argv) {
     }
 
     if (test_type_required_type && test_type_required_json) {
-        auto j = parse_json_safely(test_type_required_json);
-        if (j.is_discarded()) {
-            std::cout << "error: invalid JSON" << std::endl;
-            return 1;
-        }
-        
+        auto j = expect_or_exit(parse_json_safely(test_type_required_json));
         std::string type_str(test_type_required_type);
         auto it = json_dispatch_map.find(type_str);
         if (it != json_dispatch_map.end()) {
             it->second(j);
-        } else {
-            std::cout << "error: unknown type " << type_str << std::endl;
-            return 1;
         }
         return 0;
     }
 
+    if (test_rebase_history_size && test_rebase_content) {
+        size_t history_size = expect_or_exit(string_toul(test_rebase_history_size));
+        RebaseEngine engine;
+        auto result = engine.parse_rebase_file(test_rebase_content, history_size);
+        nlohmann::json output = expect_or_exit(result);
+        std::cout << output.dump(2) << std::endl;
+        return 0;
+    }
+
+    if (test_build_rebase_json) {
+        auto j = expect_or_exit(parse_json_safely(test_build_rebase_json));
+        auto history_result = expect_or_exit(json_to<std::vector<Message>>(j));
+        HistoryEditor editor;
+        std::string content = editor.build_rebase_content(history_result);
+        std::cout << content;
+        return 0;
+    }
+
+    if (test_apply_rebase_lines && test_apply_rebase_history) {
+        auto lines_j = expect_or_exit(parse_json_safely(test_apply_rebase_lines));
+        auto lines_result = expect_or_exit(json_to<std::vector<RebaseLine>>(lines_j));
+        auto history_j = expect_or_exit(parse_json_safely(test_apply_rebase_history));
+        auto history_result = expect_or_exit(json_to<std::vector<Message>>(history_j));
+        RebaseEngine engine;
+        auto result = engine.apply_rebase(lines_result, history_result);
+        nlohmann::json output = expect_or_exit(result);
+        std::cout << output.dump(2) << std::endl;
+        return 0;
+    }
+
+   if (test_edit_prompt_json && test_edit_prompt_index && test_edit_prompt_role) {
+        auto j = expect_or_exit(parse_json_safely(test_edit_prompt_json));
+        auto history_result = expect_or_exit(json_to<std::vector<Message>>(j));
+        size_t edit_index = expect_or_exit(string_toul(test_edit_prompt_index));
+        const std::string role = test_edit_prompt_role;
+        HistoryEditor editor;
+        std::string prompt = editor.build_edit_prompt(history_result, edit_index, role);
+        std::cout << prompt;
+        return 0;
+    }
+
     if (test_base64_encode) {
         std::cout << base64_encode(test_base64_encode) << std::endl;
         return 0;
@@ -463,11 +648,7 @@ int main(int argc, char** argv) {
         }
 
         std::string file_content((std::istreambuf_iterator<char>(file)), std::istreambuf_iterator<char>());
-        nlohmann::json data = parse_json_safely(file_content);
-        if (data.is_discarded()) {
-            std::cerr << "Error parsing JSON: invalid JSON" << "\n";
-            return 1;
-        }
+        auto data = expect_or_exit(parse_json_safely(file_content));
         print_history(std::cout, data);
         return 0;
     }
@@ -513,33 +694,28 @@ int main(int argc, char** argv) {
         }
     }
 
-    Harness harness(create_client(use_debug), agent_tools, {.max_tool_iterations = 1000, .verbose = true});
+    Harness harness(create_client(use_debug), agent_tools, harnessConfig);
 
     if (!start_new)
     {
-        std::ifstream old_harness_state(file_to_open);
-        std::string file_content((std::istreambuf_iterator<char>(old_harness_state)), std::istreambuf_iterator<char>());
-        nlohmann::json data = parse_json_safely(file_content);
-        if (data.is_discarded()) {
-            std::cerr << "Failed to parse chat file: invalid JSON" << std::endl;
-            return 1;
-        }
-        auto state_result = json_to<HarnessState>(data);
-        if (!state_result) {
-            std::cerr << "Failed to parse chat file: " << state_result.error() << std::endl;
-            return 1;
-        }
-        HarnessState state_data = state_result.value();
-
+        HarnessState state_data = read_harness_state_or_exit(file_to_open);
+        harnessConfig = state_data.config;
         harness = Harness(create_client(use_debug), master_registery, state_data);
     }
 
-    std::cerr << "drharness v0.2 | model: " << cfg.model
-              << " | " << cfg.base_url << "\n\n";
+    std::cerr << "drharness v0.3 | model: " << harnessConfig.model
+              << " | " << harnessConfig.base_url << "\n\n";
 
     std::string final_prompt;
 
-    if (prompt) {
+    if (prompt_file) {
+        std::ifstream pf(prompt_file);
+        if (!pf.is_open()) {
+            std::cerr << "Error: Cannot open prompt file " << prompt_file << "\n";
+            return 1;
+        }
+        final_prompt = std::string((std::istreambuf_iterator<char>(pf)), std::istreambuf_iterator<char>());
+    } else if (prompt) {
         final_prompt = prompt;
     } else {
         final_prompt = expect_or_exit(run_editor(harness.serialize(), full_history));

src/ollama_client.cpp

@@ -4,10 +4,12 @@
 #include <string>
 #include <iostream>
 #include <expected>
+#include "utility.hpp"
 
 static int ollama_clients = 0;
 
-OllamaClient::OllamaClient(OllamaConfig config) : config_(std::move(config)) {
+OllamaClient::OllamaClient(std::string base_url, std::string model, OllamaConfig config) 
+    : base_url_(std::move(base_url)), model_(std::move(model)), config_(std::move(config)) {
     if (ollama_clients == 0)
         curl_global_init(CURL_GLOBAL_DEFAULT);
 
@@ -37,6 +39,26 @@ size_t OllamaClient::write_cb(char* ptr, size_t size, size_t nmemb, void* userda
     return size * nmemb;
 }
 
+static int xferinfo_cb(void* cookie,
+                       curl_off_t dltotal,
+                       curl_off_t dlnow,
+                       curl_off_t ultotal,
+                       curl_off_t ulnow)
+{
+    (void)cookie;
+    (void)dltotal;
+    (void)dlnow;
+    (void)ultotal;
+    (void)ulnow;
+
+    if (ctrlc_request)
+    {
+        return 1; // Cancel
+    }
+
+    return 0; // Continue transfer
+}
+
 std::expected<std::string, std::string> OllamaClient::http_request(const std::string& full_url, 
                                        const std::string& method, 
                                        const std::string& body,
@@ -70,6 +92,8 @@ std::expected<std::string, std::string> OllamaClient::http_request(const std::st
         headers = curl_slist_append(headers, "Content-Type: application/json");
 
     curl_easy_setopt(curl, CURLOPT_URL, url.c_str());
+    curl_easy_setopt(curl, CURLOPT_XFERINFOFUNCTION, xferinfo_cb);
+    curl_easy_setopt(curl, CURLOPT_NOPROGRESS, 0L);
     curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION, write_cb);
     curl_easy_setopt(curl, CURLOPT_WRITEDATA, &response_buf);
     curl_easy_setopt(curl, CURLOPT_HTTPHEADER, headers);
@@ -98,7 +122,7 @@ std::expected<std::string, std::string> OllamaClient::http_request(const std::st
 nlohmann::json OllamaClient::build_body(const std::vector<Message>& messages,
                                         const std::vector<Tool>& tools) const {
     nlohmann::json body = {
-        {"model",    config_.model},
+        {"model",    model_},
         {"stream",   true},
         {"messages", messages},
         {"stream_options", {
@@ -119,7 +143,7 @@ std::expected<Message, std::string> OllamaClient::chat(const std::vector<Message
                                 const std::vector<Tool>& tools) const {
     nlohmann::json body = build_body(messages, tools);
     std::string body_str = body.dump();
-    std::string url = config_.base_url + "/v1/chat/completions";
+    std::string url = base_url_ + "/v1/chat/completions";
 
     Message total{};
     Usage usage{};
@@ -137,12 +161,13 @@ std::expected<Message, std::string> OllamaClient::chat(const std::vector<Message
             }
 
             //std::cerr << new_packet << std::endl;
-            auto j = parse_json_safely(new_packet.substr(6));
-            if (j.is_discarded())
+            auto j_result = parse_json_safely(new_packet.substr(6));
+            if (!j_result)
             {
                 std::cerr << "\nDiscarding: " << new_packet << std::endl;
                 continue;
             }
+            auto j = j_result.value();
 
             auto chat_response = json_to<ChatResponse>(j);
             if (!chat_response.has_value())
@@ -226,25 +251,26 @@ std::expected<Message, std::string> OllamaClient::chat(const std::vector<Message
 }
 
 std::expected<std::vector<std::string>, std::string> OllamaClient::list_models() const {
-    std::string url = config_.base_url + "/api/tags";
+    std::string url = base_url_ + "/v1/models";
     auto response_result = http_request(url, "GET", "");
     if (!response_result) {
         return std::unexpected(response_result.error());
     }
     std::string response_str = response_result.value();
 
-    auto j = parse_json_safely(response_str);
-    if (j.is_discarded()) {
-        return std::unexpected("JSON parse error: input was discarded");
+    auto j_result = parse_json_safely(response_str);
+    if (!j_result) {
+        return std::unexpected("JSON parse error: " + j_result.error());
     }
+    auto j = j_result.value();
     auto result = json_to<ModelListResponse>(j);
     if (!result) {
         return std::unexpected(std::string("JSON parse error: ") + result.error());
     }
     auto resp = result.value();
     std::vector<std::string> models;
-    for (const auto& m : resp.models) {
-        models.push_back(m.name);
+    for (const auto& m : resp.data) {
+        models.push_back(m.id);
     }
     return models;
 }

src/rebase_engine.cpp

@@ -0,0 +1,322 @@
+#include "rebase_engine.hpp"
+#include "json_types.hpp"
+#include "utility.hpp"
+#include <sstream>
+#include <algorithm>
+#include <unordered_set>
+
+static std::vector<std::string> split_lines(const std::string& content) {
+    std::vector<std::string> lines;
+    std::istringstream stream(content);
+    std::string line;
+    while (std::getline(stream, line)) {
+        lines.push_back(line);
+    }
+    return lines;
+}
+
+static std::string format_error(size_t line_num, const std::string& message) {
+    return "line " + std::to_string(line_num + 1) + ": " + message;
+}
+
+static std::expected<RebaseCommand, std::string> parse_command(const std::string& cmd) {
+    if (cmd == "pick") return RebaseCommand::Pick;
+    if (cmd == "edit") return RebaseCommand::Edit;
+    if (cmd == "drop") return RebaseCommand::Drop;
+    return std::unexpected("invalid command: '" + cmd + "'");
+}
+
+std::expected<std::vector<RebaseLine>, std::string> RebaseEngine::parse_rebase_file(
+    const std::string& content, size_t history_size) const {
+    
+    std::vector<RebaseLine> lines;
+    std::unordered_set<size_t> seen_indices;
+    
+    auto file_lines = split_lines(content);
+    
+    for (size_t line_num = 0; line_num < file_lines.size(); ++line_num) {
+        std::string line = trim(file_lines[line_num]);
+        
+        // Skip empty lines (treated as drops) and comment lines
+        if (line.empty() || line[0] == '#') {
+            continue;
+        }
+        
+        // Parse: <command> <index> [<role>] <content...>
+        std::istringstream iss(line);
+        std::string cmd_str;
+        std::string index_str;
+        std::string role_str;
+        
+        if (!(iss >> cmd_str)) {
+            return std::unexpected(format_error(line_num, "missing command"));
+        }
+        
+        if (!(iss >> index_str)) {
+            return std::unexpected(format_error(line_num, "missing index"));
+        }
+        
+        if (!(iss >> role_str)) {
+            return std::unexpected(format_error(line_num, "missing role"));
+        }
+        
+        // Parse command
+        auto cmd_result = parse_command(cmd_str);
+        if (!cmd_result) {
+            return std::unexpected(format_error(line_num, cmd_result.error()));
+        }
+        
+        // Parse index
+        size_t index;
+        auto index_result = string_toul(index_str);
+        if (!index_result) {
+            return std::unexpected(format_error(line_num, "invalid index: '" + index_str + "'"));
+        }
+        index = index_result.value();
+        
+        // Validate index is in range
+        if (index >= history_size) {
+            return std::unexpected(format_error(line_num, "index " + std::to_string(index) + " out of range (history size: " + std::to_string(history_size) + ")"));
+        }
+        
+        // Check for duplicate indices
+        if (seen_indices.count(index)) {
+            return std::unexpected(format_error(line_num, "duplicate index " + std::to_string(index)));
+        }
+        
+        // Parse and validate role
+        // Role must be in format [user], [assistant], or [tool]
+        if (role_str.size() < 3 || role_str.front() != '[' || role_str.back() != ']') {
+            return std::unexpected(format_error(line_num, "malformed role: '" + role_str + "'"));
+        }
+        
+        std::string role_content = role_str.substr(1, role_str.size() - 2);
+        if (role_content != "user" && role_content != "assistant" && role_content != "tool") {
+            return std::unexpected(format_error(line_num, "invalid role: '" + role_str + "'"));
+        }
+        
+        // Valid line - add to results
+        seen_indices.insert(index);
+        lines.push_back({cmd_result.value(), index, role_content, ""});
+    }
+    
+    return lines;
+}
+
+std::string HistoryEditor::get_first_line(const std::string& content) const {
+    auto pos = content.find('\n');
+    if (pos == std::string::npos) {
+        return content;
+    }
+    return content.substr(0, pos);
+}
+
+std::string HistoryEditor::get_message_display(const Message& msg) const {
+    // Check for assistant messages with tool_calls only
+    if (msg.role == "assistant" && !msg.tool_calls.empty()) {
+        if (msg.tool_calls.size() == 1) {
+            return "[tool_call: " + msg.tool_calls[0].function.name + "]";
+        } else {
+            // Multiple tool calls - show first one
+            return "[tool_call: " + msg.tool_calls[0].function.name + " ...]";
+        }
+    }
+
+    // Get content from message
+    std::string content = msg.get_content_string();
+
+    // If no content, show (empty)
+    if (content.empty()) {
+        return "(empty)";
+    }
+
+    // Return first line of content
+    return get_first_line(content);
+}
+
+std::string HistoryEditor::build_rebase_content(const std::vector<Message>& history) const {
+    std::ostringstream oss;
+
+    // Header comment block
+    oss << "# History Rebase Editor\n";
+    oss << "#\n";
+    oss << "# Commands:\n";
+    oss << "#   pick  - keep the message unchanged\n";
+    oss << "#   edit  - modify the message content and/or role\n";
+    oss << "#   drop  - remove the message from history\n";
+    oss << "#\n";
+    oss << "# Each line format: <command> <index> [<role>] <display_content>\n";
+    oss << "#   - index: 0-based message index\n";
+    oss << "#   - role: [user], [assistant], or [tool] (editable)\n";
+    oss << "#   - display_content: first line of message (display only, not editable here)\n";
+    oss << "#\n";
+    oss << "# Lines starting with # are comments and will be ignored.\n";
+    oss << "# Empty lines are treated as drops.\n";
+    oss << "#\n";
+    oss << "-- >8 --\n";
+
+    // Format each message
+    for (size_t i = 0; i < history.size(); ++i) {
+        const auto& msg = history[i];
+        std::string display = get_message_display(msg);
+        oss << "pick " << i << " [" << msg.role << "] " << display << "\n";
+    }
+
+    return oss.str();
+}
+
+std::string HistoryEditor::build_edit_prompt(const std::vector<Message>& history,
+                                               size_t edit_index,
+                                               const std::string& role) const {
+    std::ostringstream oss;
+
+    // Show previous message as context
+    if (edit_index == 0) {
+        oss << "# (no previous message)\n";
+    } else {
+        const auto& prev_msg = history[edit_index - 1];
+        oss << "# Previous message:\n\n";
+        oss << "[" << (edit_index - 1) << "] " << prev_msg.role << ": " << prev_msg.content << "\n";
+    }
+
+    // Edit prompt
+    oss << "\nEdit message " << edit_index << " (role: " << role << ") below:\n";
+
+    // Separator - editable content goes after this
+    oss << "\n-- >8 --\n";
+
+    // Current content (editable)
+    if (edit_index < history.size()) {
+        oss << history[edit_index].content;
+    }
+
+    return oss.str();
+}
+
+std::expected<std::vector<Message>, std::string> RebaseEngine::apply_rebase(
+    const std::vector<RebaseLine>& lines,
+    const std::vector<Message>& history) const {
+    
+    // Build a map from index to RebaseLine for quick lookup
+    std::unordered_map<size_t, const RebaseLine*> command_map;
+    for (const auto& line : lines) {
+        command_map[line.index] = &line;
+    }
+    
+    std::vector<Message> result;
+    
+    // Process messages in forward order (lowest index first)
+    for (size_t i = 0; i < history.size(); ++i) {
+        auto it = command_map.find(i);
+        
+        if (it == command_map.end()) {
+            // No command for this index - implicit drop (message is removed)
+            // Special case: if no commands at all were parsed, keep all messages
+            if (command_map.empty()) {
+                result.push_back(history[i]);
+            }
+            // Otherwise, the message is dropped (do nothing)
+        } else {
+            const auto& cmd = it->second;
+            
+            switch (cmd->command) {
+                case RebaseCommand::Pick: {
+                    // Keep message unchanged
+                    result.push_back(history[i]);
+                    break;
+                }
+                case RebaseCommand::Drop: {
+                    // Remove message from result (don't add it)
+                    break;
+                }
+                case RebaseCommand::Edit: {
+                    // Create a copy and update role and content
+                    Message edited = history[i];
+                    edited.role = cmd->role;
+                    // Update content only if it was explicitly edited
+                    if (cmd->content_edited) {
+                        edited.content = cmd->content;
+                    }
+                    result.push_back(std::move(edited));
+                    break;
+                }
+            }
+        }
+    }
+    
+    return result;
+}
+
+std::expected<std::vector<Message>, std::string> HistoryEditor::edit_history_interactive(
+    const std::vector<Message>& history) const {
+    
+    // Step 1: Build rebase content and spawn editor
+    std::string rebase_content = build_rebase_content(history);
+    auto editor_result = spawn_editor(rebase_content);
+    if (!editor_result) {
+        return std::unexpected("Failed to run rebase editor: " + editor_result.error());
+    }
+    std::string rebase_output = editor_result.value();
+    
+    // Step 2: Parse and validate rebase commands
+    RebaseEngine engine;
+    auto parse_result = engine.parse_rebase_file(rebase_output, history.size());
+    if (!parse_result) {
+        return std::unexpected("Rebase parse error: " + parse_result.error());
+    }
+    std::vector<RebaseLine> lines = parse_result.value();
+    
+    // Step 3: For each edit command (in forward order), build edit prompt and spawn editor
+    // First, sort lines by index to ensure forward order
+    std::sort(lines.begin(), lines.end(), [](const RebaseLine& a, const RebaseLine& b) {
+        return a.index < b.index;
+    });
+    
+    // Collect edit indices first (forward order)
+    std::vector<size_t> edit_indices;
+    for (const auto& line : lines) {
+        if (line.command == RebaseCommand::Edit) {
+            edit_indices.push_back(line.index);
+        }
+    }
+    
+    // For each edit, spawn editor to get new content
+    for (size_t edit_idx : edit_indices) {
+        // Find the RebaseLine for this index
+        const RebaseLine* edit_line = nullptr;
+        for (const auto& line : lines) {
+            if (line.index == edit_idx && line.command == RebaseCommand::Edit) {
+                edit_line = &line;
+                break;
+            }
+        }
+        if (!edit_line) continue;
+        
+        // Build edit prompt with context
+        std::string prompt = build_edit_prompt(history, edit_idx, edit_line->role);
+        
+        // Spawn editor to get new content
+        auto edit_result = spawn_editor(prompt);
+        if (!edit_result) {
+            return std::unexpected("Failed to run edit editor for message " + 
+                                   std::to_string(edit_idx) + ": " + edit_result.error());
+        }
+        
+        // Update the RebaseLine with the new content
+        for (auto& line : lines) {
+            if (line.index == edit_idx && line.command == RebaseCommand::Edit) {
+                line.content = edit_result.value();
+                line.content_edited = true;  // Mark that content was explicitly edited
+                break;
+            }
+        }
+    }
+    
+    // Step 4: Apply all changes
+    auto apply_result = engine.apply_rebase(lines, history);
+    if (!apply_result) {
+        return std::unexpected("Failed to apply rebase: " + apply_result.error());
+    }
+    
+    return apply_result.value();
+}

src/utility.cpp

@@ -0,0 +1,104 @@
+#include "utility.hpp"
+#include <fstream>
+#include <cstdlib>
+#include <filesystem>
+#include <random>
+#include <cctype>
+
+std::string trim(const std::string& s) {
+    auto start = s.find_first_not_of(" \t\n\r");
+    if (start == std::string::npos) return "";
+    auto end = s.find_last_not_of(" \t\n\r");
+    return s.substr(start, end - start + 1);
+}
+
+std::string random_string(size_t length) {
+    static const char chars[] =
+        "abcdefghijklmnopqrstuvwxyz"
+        "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
+        "0123456789";
+
+    static std::random_device rd;
+    static std::mt19937 gen(rd());
+    static std::uniform_int_distribution<> dis(0, sizeof(chars) - 2);
+
+    std::string s;
+    for (size_t i = 0; i < length; ++i) {
+        s += chars[dis(gen)];
+    }
+    return s;
+}
+
+std::string random_file_name(size_t length) {
+    std::filesystem::path temp_dir = std::filesystem::temp_directory_path();
+    std::filesystem::path temp_file = temp_dir / ("tmp_" + random_string(length));
+    return temp_file.string();
+}
+
+std::expected<std::string, std::string> spawn_editor(const std::string& initial_content) {
+    auto file_path = random_file_name();
+    auto cleanup = Defer([&](){
+        std::filesystem::remove(file_path);
+    });
+
+    {
+        std::ofstream file(file_path);
+        if (!file.is_open()) {
+            return std::unexpected("Failed to open temp file for editor.");
+        }
+        file << initial_content;
+        file.close();
+
+        const char* editor_env = getenv("VISUAL");
+        if (!editor_env) editor_env = getenv("EDITOR");
+        if (!editor_env) {
+            return std::unexpected("Neither $VISUAL nor $EDITOR is set.");
+        }
+
+        std::string cmd = std::string(editor_env) + " " + file_path;
+        int res = system(cmd.c_str());
+        if (res != 0) {
+            return std::unexpected("Editor exited with error.");
+        }
+
+        std::ifstream infile(file_path);
+        std::string content((std::istreambuf_iterator<char>(infile)), std::istreambuf_iterator<char>());
+        
+        const std::string seperator = "\n-- >8 --\n";
+        size_t pos = content.find(seperator);
+        if (pos == std::string::npos) {
+            return std::unexpected("Scissors not found.");
+        }
+        
+        size_t last_pos = std::string::npos;
+        while (pos != std::string::npos) {
+            last_pos = pos;
+            pos = content.find(seperator, pos + 1);
+        }
+        
+        std::string extracted = content.substr(last_pos + seperator.size());
+        extracted = trim(extracted);
+        
+        return extracted;
+    }
+}
+
+std::expected<unsigned long, std::string> string_toul(std::string const& str) {
+    size_t pos = 0;
+    try {
+        unsigned long result = std::stoul(str, &pos);
+        
+        // Check if the entire string was consumed
+        while (pos < str.size() && std::isspace(static_cast<unsigned char>(str[pos]))) {
+            pos++;
+        }
+        
+        if (pos != str.size()) {
+            return std::unexpected("trailing characters after number");
+        }
+        
+        return result;
+    } catch (const std::exception& e) {
+        return std::unexpected(e.what());
+    }
+}

tests/CMakeLists.txt

@@ -19,11 +19,14 @@ add_dr_test(test_tool_read_file)
 add_dr_test(test_tool_list_files)
 add_dr_test(test_tool_grep)
 add_dr_test(test_tool_shell)
+add_dr_test(test_tool_shell_ctrlc)
+add_dr_test(test_ctrlc_after_tool)
 add_dr_test(test_tool_read_multi_modal)
 
 # Add feature validation tests
 add_dr_test(test_tool_restrictions)
 add_dr_test(test_n_flag)
+add_dr_test(test_prompt_file)
 add_dr_test(test_load_chat)
 add_dr_test(process_reasoning)
 add_dr_test(test_skill_scanner)
@@ -55,6 +58,90 @@ add_dr_test(test_json_variant)
 # Add expected error handling tests
 add_dr_test(test_expected_errors)
 
+# Add rebase parser tests (split into individual tests)
+add_dr_test(test_rebase_parser_all_commands)
+add_dr_test(test_rebase_parser_with_comments)
+add_dr_test(test_rebase_parser_with_empty_lines)
+add_dr_test(test_rebase_parser_invalid_command)
+add_dr_test(test_rebase_parser_index_out_of_range)
+add_dr_test(test_rebase_parser_duplicate_indices)
+add_dr_test(test_rebase_parser_malformed_role)
+add_dr_test(test_rebase_parser_invalid_role_value)
+add_dr_test(test_rebase_parser_first_error_returned)
+add_dr_test(test_rebase_parser_empty_rebase)
+add_dr_test(test_rebase_parser_only_comments)
+add_dr_test(test_rebase_parser_all_valid_roles)
+add_dr_test(test_rebase_parser_invalid_command_digit)
+add_dr_test(test_rebase_parser_invalid_index)
+add_dr_test(test_rebase_parser_missing_role)
+add_dr_test(test_rebase_parser_whitespace_variations)
+add_dr_test(test_rebase_parser_comments_and_empty_lines_mixed)
+
+# Add rebase content formatting tests (split into individual tests)
+add_dr_test(test_rebase_content_basic_messages)
+add_dr_test(test_rebase_content_multiline_content)
+add_dr_test(test_rebase_content_tool_call_display)
+add_dr_test(test_rebase_content_tool_message)
+add_dr_test(test_rebase_content_empty_message)
+add_dr_test(test_rebase_content_header_comments)
+add_dr_test(test_rebase_content_multiple_tool_calls)
+add_dr_test(test_rebase_content_message_indices)
+add_dr_test(test_rebase_content_scissors_line)
+add_dr_test(test_rebase_content_special_characters)
+add_dr_test(test_rebase_content_unicode_characters)
+add_dr_test(test_rebase_content_very_long_content)
+
+# Add rebase apply tests (split into individual tests)
+add_dr_test(test_rebase_apply_pick_preserves)
+add_dr_test(test_rebase_apply_drop_removes)
+add_dr_test(test_rebase_apply_edit_updates_role)
+add_dr_test(test_rebase_apply_edit_preserves_fields)
+add_dr_test(test_rebase_apply_implicit_pick)
+add_dr_test(test_rebase_apply_mixed_commands)
+add_dr_test(test_rebase_apply_empty_lines_preserves)
+add_dr_test(test_rebase_apply_lenient_orphaned_results)
+add_dr_test(test_rebase_apply_edit_preserves_tool_calls)
+add_dr_test(test_rebase_apply_deleting_lines_drops)
+add_dr_test(test_rebase_apply_all_lines_deleted_keeps)
+
+# Add edit prompt tests (split from test_edit_prompt.sh)
+add_dr_test(test_edit_prompt_with_context)
+add_dr_test(test_edit_prompt_first_message)
+add_dr_test(test_edit_prompt_separator_placement)
+add_dr_test(test_edit_prompt_multiline_context)
+add_dr_test(test_edit_prompt_tool_role)
+add_dr_test(test_edit_prompt_middle_message)
+
+# Add edit history flag tests (split from test_edit_history_flag.sh)
+add_dr_test(test_edit_history_flag_requires_F_short)
+add_dr_test(test_edit_history_flag_requires_F_long)
+add_dr_test(test_edit_history_flag_empty_file_short)
+add_dr_test(test_edit_history_flag_no_history)
+add_dr_test(test_edit_history_flag_empty_file_long)
+
+# Add edit history integration tests (split into individual tests)
+add_dr_test(test_edit_history_pick_all_messages)
+add_dr_test(test_edit_history_no_edits_unchanged)
+add_dr_test(test_edit_history_drop_single_message)
+add_dr_test(test_edit_history_edit_message_content)
+add_dr_test(test_edit_history_change_message_role)
+add_dr_test(test_edit_history_delete_line_equals_drop)
+add_dr_test(test_edit_history_edit_first_message)
+add_dr_test(test_edit_history_prompt_noop)
+add_dr_test(test_edit_history_empty_history_rejected)
+add_dr_test(test_edit_history_invalid_command_error)
+add_dr_test(test_edit_history_invalid_index_error)
+add_dr_test(test_edit_history_invalid_role_error)
+add_dr_test(test_edit_history_multiple_edits)
+add_dr_test(test_edit_history_then_continue)
+add_dr_test(test_edit_history_remove_all_content)
+add_dr_test(test_edit_history_multiple_consecutive_edits)
+add_dr_test(test_edit_history_special_characters)
+add_dr_test(test_edit_history_very_long_content)
+add_dr_test(test_edit_history_assistant_with_tool_calls)
+add_dr_test(test_edit_history_drop_all_messages)
+add_dr_test(test_edit_history_unicode_characters)
+
 # Add base64 encoding tests
 add_dr_test(test_base64_encode)
 
@@ -75,3 +162,6 @@ add_dr_test(test_print_history_multimodal)
 
 # Add encoding tests
 add_dr_test(test_encoding)
+
+# Add harness config tests (base_url and model in HarnessConfig)
+add_dr_test(test_harness_config)

tests/mock_editor.sh

@@ -2,6 +2,33 @@
 # $1 is the file path passed to the editor
 FILE=$1
 
+# Support sequential invocations using a counter file
+# Track invocations in $TEST_TEMP_DIR/.editor_invocation_count
+# Copy from $TEST_TEMP_DIR/.editor_input_$INVOCATION to the editor file
+if [ -n "$TEST_TEMP_DIR" ] && [ -f "$TEST_TEMP_DIR/.editor_input_0" ]; then
+    # Initialize counter file if it doesn't exist
+    if [ ! -f "$TEST_TEMP_DIR/.editor_invocation_count" ]; then
+        echo "0" > "$TEST_TEMP_DIR/.editor_invocation_count"
+    fi
+    
+    # Read current invocation count
+    INVOCATION=$(cat "$TEST_TEMP_DIR/.editor_invocation_count")
+    
+    # Check if there's a staged input file for this invocation
+    STAGED_INPUT="$TEST_TEMP_DIR/.editor_input_$INVOCATION"
+    if [ -f "$STAGED_INPUT" ]; then
+        # Copy staged content to the editor file
+        cp "$STAGED_INPUT" "$FILE"
+    fi
+    
+    # Increment the counter for next invocation
+    echo $((INVOCATION + 1)) > "$TEST_TEMP_DIR/.editor_invocation_count"
+    
+    # For inspecting by tests
+    cat $FILE > hist.md
+    exit 0
+fi
+
 # For inspecting by tests
 cat $FILE > hist.md
 

tests/process_reasoning.sh

@@ -8,12 +8,12 @@ $DRHARNESS -n -F reasoning_content.json -m "Test" -d < reasoning_content
 
 
 
-if [[ "$(cat reasoning.json | jq .history[1].reasoning -r)" != "I should do X because Y" ]]; then
-    cat reasoning.json | jq
+if [[ "$(dr_reasoning reasoning.json 1)" != "I should do X because Y" ]]; then
+    echo "reasoning mismatch"
     exit 1
 fi
-if [[ "$(cat reasoning_content.json | jq .history[1].reasoning_content -r)" != "I should do X because Y" ]]; then
-    cat reasoning_content.json | jq
+if [[ "$(dr_reasoning_content reasoning_content.json 1)" != "I should do X because Y" ]]; then
+    echo "reasoning_content mismatch"
     exit 1
 fi
 

tests/run_test.sh

@@ -144,6 +144,223 @@ function dr_result() {
 }
 export -f dr_result
 
+function dr_role() {
+    local chat_file="$1"
+    local entry="$2"
+    jq -r --arg e "$entry" '.history[$e | tonumber].role' "$chat_file" 2>/dev/null || echo ""
+}
+export -f dr_role
+
+function dr_reasoning() {
+    local chat_file="$1"
+    local entry="$2"
+    jq -r --arg e "$entry" '.history[$e | tonumber].reasoning // ""' "$chat_file" 2>/dev/null || echo ""
+}
+export -f dr_reasoning
+
+function dr_reasoning_content() {
+    local chat_file="$1"
+    local entry="$2"
+    jq -r --arg e "$entry" '.history[$e | tonumber].reasoning_content // ""' "$chat_file" 2>/dev/null || echo ""
+}
+export -f dr_reasoning_content
+
+# Create a chat file with history messages
+# Usage: create_chat_file <filename> <role> <content> [role content ...]
+# Example: create_chat_file chat.json user "hello" assistant "hi there"
+function create_chat_file() {
+    local file="$1"
+    shift
+    create_history_json "$@" | jq '{history: ., tools: []}' > "$file"
+}
+export -f create_chat_file
+
+# Create a JSON array of history messages (for -T flag testing)
+# Usage: create_history_json role content [role content ...]
+# Example: create_history_json user "hello" assistant "hi there"
+function create_history_json() {
+    local history="[]"
+    while [ $# -gt 0 ]; do
+        local role="$1"
+        local content="$2"
+        shift 2
+        history=$(echo "$history" | jq --arg r "$role" --arg c "$content" '. + [{"role": $r, "content": $c}]')
+    done
+    echo "$history"
+}
+export -f create_history_json
+
+# Helper functions for rebase parser testing
+
+# Get number of lines in rebase JSON output
+function rebase_line_count() {
+    local output="$1"
+    echo "$output" | jq 'length'
+}
+export -f rebase_line_count
+
+# Get command at position in rebase JSON output
+function rebase_command_at() {
+    local output="$1"
+    local pos="$2"
+    echo "$output" | jq -r ".[$pos | tonumber].command"
+}
+export -f rebase_command_at
+
+# Get index at position in rebase JSON output
+function rebase_index_at() {
+    local output="$1"
+    local pos="$2"
+    echo "$output" | jq -r ".[$pos | tonumber].index"
+}
+export -f rebase_index_at
+
+# Get role at position in rebase JSON output
+function rebase_role_at() {
+    local output="$1"
+    local pos="$2"
+    echo "$output" | jq -r ".[$pos | tonumber].role"
+}
+export -f rebase_role_at
+
+# Helper functions for rebase apply testing
+
+# Create a rebase line JSON object
+# Usage: create_rebase_line <command> <index> <role>
+# Example: create_rebase_line pick 0 user
+function create_rebase_line() {
+    local cmd="$1"
+    local idx="$2"
+    local role="$3"
+    jq -n --arg c "$cmd" --argjson i "$idx" --arg r "$role" '{"command": $c, "index": $i, "role": $r}'
+}
+export -f create_rebase_line
+
+# Create a JSON array of rebase lines
+# Usage: create_rebase_lines [command index role ...]
+# Example: create_rebase_lines pick 0 user drop 1 assistant edit 2 tool
+function create_rebase_lines() {
+    local lines="[]"
+    while [ $# -gt 0 ]; do
+        local cmd="$1"
+        local idx="$2"
+        local role="$3"
+        shift 3
+        local line
+        line=$(create_rebase_line "$cmd" "$idx" "$role")
+        lines=$(echo "$lines" | jq --argjson l "$line" '. + [$l]')
+    done
+    echo "$lines"
+}
+export -f create_rebase_lines
+
+# Add a field to a message JSON object
+function add_message_field() {
+    local msg="$1"
+    local field="$2"
+    local value="$3"
+    case "$field" in
+        reasoning)
+            echo "$msg" | jq --arg v "$value" '. + {reasoning: $v}'
+            ;;
+        reasoning_content)
+            echo "$msg" | jq --arg v "$value" '. + {reasoning_content: $v}'
+            ;;
+        tool_call_id)
+            echo "$msg" | jq --arg v "$value" '. + {tool_call_id: $v}'
+            ;;
+        tool_call)
+            local call_id="$3"
+            local call_name="$4"
+            local call_args="$5"
+            echo "$msg" | jq --arg id "$call_id" --arg name "$call_name" --arg args "$call_args" \
+                '.tool_calls = ((.tool_calls // []) + [{"index": 0, "id": $id, "type": "function", "function": {"name": $name, "arguments": $args}}])'
+            ;;
+        *)
+            echo "ERROR: Unknown field '$field'" >&2
+            return 1
+            ;;
+    esac
+}
+export -f add_message_field
+
+# Create a message JSON object with optional fields
+# Usage: create_message_json <role> <content> [field value ...]
+# Supported fields: reasoning, reasoning_content, tool_call_id, tool_call
+# tool_call takes 3 args: tool_call <id> <name> <arguments>
+# Example: create_message_json user "hello"
+# Example: create_message_json user "hello" reasoning "thought" tool_call_id "call_123"
+# Example: create_message_json assistant "" tool_call "call_1" "test" "{}"
+function create_message_json() {
+    local role="$1"
+    local content="$2"
+    shift 2
+
+    local msg
+    msg=$(jq -n --arg r "$role" --arg c "$content" '{"role": $r, "content": $c}')
+
+    while [ $# -gt 0 ]; do
+        local field="$1"
+        if [ "$field" = "tool_call" ]; then
+            local call_id="$2"
+            local call_name="$3"
+            local call_args="$4"
+            shift 4
+            msg=$(add_message_field "$msg" "$field" "$call_id" "$call_name" "$call_args")
+        else
+            local value="$2"
+            shift 2
+            msg=$(add_message_field "$msg" "$field" "$value")
+        fi
+    done
+    echo "$msg"
+}
+export -f create_message_json
+
+# Check if a string is a known message field name
+function is_message_field() {
+    case "$1" in
+        reasoning|reasoning_content|tool_call_id|tool_call) return 0 ;;
+        *) return 1 ;;
+    esac
+}
+export -f is_message_field
+
+# Create a JSON array of messages
+# Usage: create_messages_json <role> <content> [field value ...] [role content ...]
+# Each message: role content [field value ...]
+# Supported fields: reasoning, reasoning_content, tool_call_id, tool_call
+# tool_call takes 3 args: tool_call <id> <name> <arguments>
+# Example: create_messages_json user "hello" assistant "hi there"
+# Example: create_messages_json user "hello" reasoning "thought" assistant "hi"
+function create_messages_json() {
+    local messages="[]"
+
+    while [ $# -gt 0 ]; do
+        local role="$1"
+        local content="$2"
+        shift 2
+
+        # Collect field-value pairs for this message into an array
+        local -a field_args=()
+        while [ $# -gt 0 ] && is_message_field "$1"; do
+            if [ "$1" = "tool_call" ]; then
+                field_args+=("$1" "$2" "$3" "$4")
+                shift 4
+            else
+                field_args+=("$1" "$2")
+                shift 2
+            fi
+        done
+
+        local msg
+        msg=$(create_message_json "$role" "$content" "${field_args[@]}")
+        messages=$(echo "$messages" | jq --argjson m "$msg" '. + [$m]')
+    done
+    echo "$messages"
+}
+export -f create_messages_json
+
 function create_state() {
     local file="$1"
     shift
@@ -238,6 +455,20 @@ function dr_content_text() {
 }
 export -f dr_content_text
 
+# Get base_url from config in chat file
+function dr_config_base_url() {
+    local chat_file="$1"
+    jq -r '.config.base_url' "$chat_file"
+}
+export -f dr_config_base_url
+
+# Get model from config in chat file
+function dr_config_model() {
+    local chat_file="$1"
+    jq -r '.config.model' "$chat_file"
+}
+export -f dr_config_model
+
 # Resolve paths relative to this script
 SCRIPT_DIR=$(cd "$(dirname "$0")" && pwd)
 PROJECT_ROOT=$(cd "${SCRIPT_DIR}/.." && pwd)

tests/test_base64_encode.sh

@@ -16,7 +16,7 @@ if [ "${OUTPUT}" != "" ]; then
 fi
 
 # Test 3: Base64 decoding
-OUTPUT=$("${DRHARNESS}" -b "SGVsbG8sIFdvcmxkIQ==")
+OUTPUT=$("${DRHARNESS}" -K "SGVsbG8sIFdvcmxkIQ==")
 if [ "${OUTPUT}" != "Hello, World!" ]; then
     echo "FAIL: base64_decode - expected 'Hello, World!', got '${OUTPUT}'"
     exit 1
@@ -25,7 +25,7 @@ fi
 # Test 4: Round-trip encoding/decoding
 ORIGINAL="Test string with special chars: !@#$%^&*()"
 ENCODED=$("${DRHARNESS}" -B "${ORIGINAL}")
-DECODED=$("${DRHARNESS}" -b "${ENCODED}")
+DECODED=$("${DRHARNESS}" -K "${ENCODED}")
 if [ "${DECODED}" != "${ORIGINAL}" ]; then
     echo "FAIL: round-trip - expected '${ORIGINAL}', got '${DECODED}'"
     exit 1

tests/test_ctrlc_after_tool.sh

@@ -0,0 +1,61 @@
+#!/bin/bash
+# test_ctrlc_after_tool.sh - Test ctrlc (SIGINT) handling between tool calls
+
+# Create mock responses: two tool calls followed by a final response
+# First tool sleeps 1 second, second tool is instant
+# We send SIGINT during the first tool's execution (at 0.5s)
+# The new ctrlc check after tool execution should catch it at 1s
+tool_call run_dangerous_shell '{"command":"sleep 1","timeout_seconds":5}' >> mock_ctrlc_between_tools.txt
+tool_call run_dangerous_shell '{"command":"echo second_tool","timeout_seconds":5}' >> mock_ctrlc_between_tools.txt
+be_done 'Final response' >> mock_ctrlc_between_tools.txt
+
+# Run drharness in background
+"${DRHARNESS}" -F chat_ctrlc.json -n -m "Test ctrlc between tools" -t run_dangerous_shell -d < "mock_ctrlc_between_tools.txt" > /dev/null 2>&1 &
+DRHARNESS_PID=$!
+
+# Send SIGINT at 0.5s - during first tool's sleep
+sleep 0.5
+kill -SIGINT $DRHARNESS_PID 2>/dev/null || true
+
+# Wait for drharness to complete (first tool finishes at 1s, then ctrlc check catches it)
+wait $DRHARNESS_PID 2>/dev/null || true
+
+# Verify the chat file exists and has the conversation state preserved
+if [ ! -f chat_ctrlc.json ]; then
+    echo "FAIL: chat_ctrlc.json was not created"
+    exit 1
+fi
+
+# Check that we have exactly 3 history entries: user message, assistant tool call, tool result
+# The second tool call should NOT have executed because ctrlc was caught after first tool
+HISTORY_LENGTH=$(history_length chat_ctrlc.json)
+
+if [ "$HISTORY_LENGTH" -ne 3 ]; then
+    echo "FAIL: Expected exactly 3 history entries (user, assistant, tool), got $HISTORY_LENGTH"
+    echo "This indicates the second tool call was not cancelled properly"
+    cat chat_ctrlc.json
+    exit 1
+fi
+
+# Verify user message is present
+USER_ROLE=$(dr_role chat_ctrlc.json 0)
+if [ "$USER_ROLE" != "user" ]; then
+    echo "FAIL: Expected user role at index 0, got $USER_ROLE"
+    exit 1
+fi
+
+# Verify assistant with tool call is present
+ASSISTANT_ROLE=$(dr_role chat_ctrlc.json 1)
+if [ "$ASSISTANT_ROLE" != "assistant" ]; then
+    echo "FAIL: Expected assistant role at index 1, got $ASSISTANT_ROLE"
+    exit 1
+fi
+
+# Verify tool result is present
+TOOL_ROLE=$(dr_role chat_ctrlc.json 2)
+if [ "$TOOL_ROLE" != "tool" ]; then
+    echo "FAIL: Expected tool role at index 2, got $TOOL_ROLE"
+    exit 1
+fi
+
+echo "PASS: ctrlc_after_tool"

tests/test_edit_history_assistant_with_tool_calls.sh

@@ -0,0 +1,103 @@
+#!/bin/bash
+# test_edit_history_assistant_with_tool_calls.sh - Editing assistant messages that contain tool_calls
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with an assistant message containing tool_calls
+cat > chat.json << 'CHATJSON'
+{
+  "history": [
+    {"role":"user","content":"hello"},
+    {"role":"assistant","content":"calling tool","tool_calls":[{"index":0,"id":"call_1","type":"function","function":{"name":"test","arguments":"{}"}}]},
+    {"role":"tool","content":"result","tool_call_id":"call_1"}
+  ],
+  "tools": []
+}
+CHATJSON
+
+# Stage the rebase editor output (edit message 1)
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+pick 0 [user] hello
+edit 1 [assistant] calling tool [tool_call: test]
+pick 2 [tool] result
+EDITEOF
+
+# Stage the edit prompt response
+cat > .editor_input_1 << 'EDITEOF'
+# Previous message:
+
+[0] user: hello
+
+Edit message 1 (role: assistant) below:
+
+-- >8 --
+edited assistant message
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: edit_assistant_with_tool_calls - expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify history still has 3 messages
+len=$(history_length chat.json)
+if [ "$len" != "3" ]; then
+    echo "FAIL: edit_assistant_with_tool_calls - expected 3 messages, got $len"
+    exit 1
+fi
+
+# Verify message 1 content was edited
+msg1_content=$(dr_result chat.json 1)
+if [ "$msg1_content" != "edited assistant message" ]; then
+    echo "FAIL: edit_assistant_with_tool_calls - message 1 content not updated correctly"
+    echo "Expected: 'edited assistant message', got: '$msg1_content'"
+    exit 1
+fi
+
+# Verify tool_calls are preserved
+if ! cat chat.json | jq -e '.history[1].tool_calls' > /dev/null 2>&1; then
+    echo "FAIL: edit_assistant_with_tool_calls - tool_calls not preserved"
+    exit 1
+fi
+
+echo "PASS: edit_assistant_with_tool_calls"

tests/test_edit_history_change_message_role.sh

@@ -0,0 +1,79 @@
+#!/bin/bash
+# test_edit_history_change_message_role.sh - Change message role
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with two messages
+create_chat_file chat.json user "hello" assistant "hi there"
+
+# Stage the rebase editor output (edit message 1 role to user)
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+pick 0 [user] hello
+edit 1 [user] hi there
+EDITEOF
+
+# Stage the edit prompt response (keep same content)
+cat > .editor_input_1 << 'EDITEOF'
+# Previous message:
+
+[0] user: hello
+
+Edit message 1 (role: user) below:
+
+-- >8 --
+hi there
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: change_message_role - expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify message 1 role was changed
+msg1_role=$(dr_role chat.json 1)
+if [ "$msg1_role" != "user" ]; then
+    echo "FAIL: change_message_role - message 1 role should be 'user', got '$msg1_role'"
+    exit 1
+fi
+
+echo "PASS: change_message_role"

tests/test_edit_history_delete_line_equals_drop.sh

@@ -0,0 +1,67 @@
+#!/bin/bash
+# test_edit_history_delete_line_equals_drop.sh - Delete line equals drop (implicit drop for messages not mentioned)
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with two messages
+create_chat_file chat.json user "hello" assistant "hi there"
+
+# Stage rebase editor output with only message 0 mentioned
+# Messages not mentioned are implicitly dropped
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+pick 0 [user] hello
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: delete_line_equals_drop - expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Messages not mentioned are implicitly dropped, so we should have 1 message
+len=$(history_length chat.json)
+if [ "$len" != "1" ]; then
+    echo "FAIL: delete_line_equals_drop - expected 1 message, got $len"
+    exit 1
+fi
+
+echo "PASS: delete_line_equals_drop"

tests/test_edit_history_drop_all_messages.sh

@@ -0,0 +1,67 @@
+#!/bin/bash
+# test_edit_history_drop_all_messages.sh - Dropping all messages results in empty history
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with two messages
+create_chat_file chat.json user "hello" assistant "hi there"
+
+# Stage the rebase editor output (drop all)
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+drop 0 [user] hello
+drop 1 [assistant] hi there
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: drop_all_messages - expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify history is empty
+len=$(history_length chat.json)
+if [ "$len" != "0" ]; then
+    echo "FAIL: drop_all_messages - expected 0 messages, got $len"
+    exit 1
+fi
+
+echo "PASS: drop_all_messages"

tests/test_edit_history_drop_single_message.sh

@@ -0,0 +1,73 @@
+#!/bin/bash
+# test_edit_history_drop_single_message.sh - Drop a single message
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with two messages
+create_chat_file chat.json user "hello" assistant "hi there"
+
+# Stage the rebase editor output (drop message 1)
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+pick 0 [user] hello
+drop 1 [assistant] hi there
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: drop_single_message - expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify message 1 was dropped
+len=$(history_length chat.json)
+if [ "$len" != "1" ]; then
+    echo "FAIL: drop_single_message - expected 1 message, got $len"
+    exit 1
+fi
+
+msg0_content=$(dr_result chat.json 0)
+if [ "$msg0_content" != "hello" ]; then
+    echo "FAIL: drop_single_message - message 0 content should be unchanged"
+    exit 1
+fi
+
+echo "PASS: drop_single_message"

tests/test_edit_history_edit_first_message.sh

@@ -0,0 +1,78 @@
+#!/bin/bash
+# test_edit_history_edit_first_message.sh - Edit first message (no context)
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with two messages
+create_chat_file chat.json user "hello" assistant "hi there"
+
+# Stage the rebase editor output (edit message 0)
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+edit 0 [user] hello
+pick 1 [assistant] hi there
+EDITEOF
+
+# Stage the edit prompt response (message 0 has no previous message)
+cat > .editor_input_1 << 'EDITEOF'
+# (no previous message)
+
+Edit message 0 (role: user) below:
+
+-- >8 --
+goodbye
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: edit_first_message - expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify message 0 was edited
+msg0_content=$(dr_result chat.json 0)
+if [ "$msg0_content" != "goodbye" ]; then
+    echo "FAIL: edit_first_message - message 0 content not updated correctly"
+    echo "Expected: 'goodbye', got: '$msg0_content'"
+    exit 1
+fi
+
+echo "PASS: edit_first_message"

tests/test_edit_history_edit_message_content.sh

@@ -0,0 +1,86 @@
+#!/bin/bash
+# test_edit_history_edit_message_content.sh - Edit message content
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with two messages
+create_chat_file chat.json user "hello" assistant "hi there"
+
+# Stage the rebase editor output (edit message 1)
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+pick 0 [user] hello
+edit 1 [assistant] hi there
+EDITEOF
+
+# Stage the edit prompt response (new content for message 1)
+cat > .editor_input_1 << 'EDITEOF'
+# Previous message:
+
+[0] user: hello
+
+Edit message 1 (role: assistant) below:
+
+-- >8 --
+edited content here
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: edit_message_content - expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify message 1 was edited
+len=$(history_length chat.json)
+if [ "$len" != "2" ]; then
+    echo "FAIL: edit_message_content - expected 2 messages, got $len"
+    exit 1
+fi
+
+msg1_content=$(dr_result chat.json 1)
+if [ "$msg1_content" != "edited content here" ]; then
+    echo "FAIL: edit_message_content - message 1 content not updated correctly"
+    echo "Expected: 'edited content here', got: '$msg1_content'"
+    exit 1
+fi
+
+echo "PASS: edit_message_content"

tests/test_edit_history_empty_history_rejected.sh

@@ -0,0 +1,43 @@
+#!/bin/bash
+# test_edit_history_empty_history_rejected.sh - Empty history rejected
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with no history
+create_chat_file chat.json
+
+# Run drharness with -b flag
+set +e
+run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -eq 0 ]; then
+    echo "FAIL: empty_history_rejected - expected non-zero exit code"
+    exit 1
+fi
+
+if ! echo "$OUTPUT" | grep -q "no history"; then
+    echo "FAIL: empty_history_rejected - expected error about no history"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: empty_history_rejected"

tests/test_edit_history_flag_empty_file_long.sh

@@ -0,0 +1,17 @@
+#!/bin/bash
+# test_edit_history_flag_empty_file_long.sh - Test that --edit-history with empty file fails
+
+# Create empty file
+> empty_chat.json
+
+set +e
+OUTPUT=$("${DRHARNESS}" --edit-history -F empty_chat.json 2>&1)
+EXIT_CODE=$?
+set -e
+
+if [ ${EXIT_CODE} -eq 0 ]; then
+    echo "FAIL: expected non-zero exit code, got 0"
+    exit 1
+fi
+
+echo "PASS: edit_history_flag_empty_file_long"

tests/test_edit_history_flag_empty_file_short.sh

@@ -0,0 +1,17 @@
+#!/bin/bash
+# test_edit_history_flag_empty_file_short.sh - Test that -b with empty file fails
+
+# Create empty file
+> empty_chat.json
+
+set +e
+OUTPUT=$("${DRHARNESS}" -b -F empty_chat.json 2>&1)
+EXIT_CODE=$?
+set -e
+
+if [ ${EXIT_CODE} -eq 0 ]; then
+    echo "FAIL: expected non-zero exit code, got 0"
+    exit 1
+fi
+
+echo "PASS: edit_history_flag_empty_file_short"

tests/test_edit_history_flag_no_history.sh

@@ -0,0 +1,23 @@
+#!/bin/bash
+# test_edit_history_flag_no_history.sh - Test that -b with no history fails
+
+# Create file with valid JSON but empty history
+create_chat_file no_history.json
+
+set +e
+OUTPUT=$("${DRHARNESS}" -b -F no_history.json 2>&1)
+EXIT_CODE=$?
+set -e
+
+if [ ${EXIT_CODE} -eq 0 ]; then
+    echo "FAIL: expected non-zero exit code, got 0"
+    exit 1
+fi
+
+if ! echo "$OUTPUT" | grep -q "no history"; then
+    echo "FAIL: expected error message about no history"
+    echo "Output was: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: edit_history_flag_no_history"

tests/test_edit_history_flag_requires_F_long.sh

@@ -0,0 +1,20 @@
+#!/bin/bash
+# test_edit_history_flag_requires_F_long.sh - Test that --edit-history without -F fails
+
+set +e
+OUTPUT=$("${DRHARNESS}" --edit-history 2>&1)
+EXIT_CODE=$?
+set -e
+
+if [ ${EXIT_CODE} -eq 0 ]; then
+    echo "FAIL: expected non-zero exit code, got 0"
+    exit 1
+fi
+
+if ! echo "$OUTPUT" | grep -q "requires -F"; then
+    echo "FAIL: expected error message about -F requirement"
+    echo "Output was: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: edit_history_flag_requires_F_long"

tests/test_edit_history_flag_requires_F_short.sh

@@ -0,0 +1,20 @@
+#!/bin/bash
+# test_edit_history_flag_requires_F_short.sh - Test that -b without -F fails
+
+set +e
+OUTPUT=$("${DRHARNESS}" -b 2>&1)
+EXIT_CODE=$?
+set -e
+
+if [ ${EXIT_CODE} -eq 0 ]; then
+    echo "FAIL: expected non-zero exit code, got 0"
+    exit 1
+fi
+
+if ! echo "$OUTPUT" | grep -q "requires -F"; then
+    echo "FAIL: expected error message about -F requirement"
+    echo "Output was: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: edit_history_flag_requires_F_short"

tests/test_edit_history_invalid_command_error.sh

@@ -0,0 +1,64 @@
+#!/bin/bash
+# test_edit_history_invalid_command_error.sh - Invalid command error
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with a message
+create_chat_file chat.json user "hello"
+
+# Stage the rebase editor output with invalid command
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+foo 0 [user] hello
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -eq 0 ]; then
+    echo "FAIL: invalid_command_error - expected non-zero exit code"
+    exit 1
+fi
+
+if ! echo "$OUTPUT" | grep -qi "invalid\|error"; then
+    echo "FAIL: invalid_command_error - expected error message"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: invalid_command_error"

tests/test_edit_history_invalid_index_error.sh

@@ -0,0 +1,64 @@
+#!/bin/bash
+# test_edit_history_invalid_index_error.sh - Invalid index error
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with one message
+create_chat_file chat.json user "hello"
+
+# Stage the rebase editor output with out-of-range index
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+pick 5 [user] hello
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -eq 0 ]; then
+    echo "FAIL: invalid_index_error - expected non-zero exit code"
+    exit 1
+fi
+
+if ! echo "$OUTPUT" | grep -qi "out of range\|error"; then
+    echo "FAIL: invalid_index_error - expected error about out of range"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: invalid_index_error"

tests/test_edit_history_invalid_role_error.sh

@@ -0,0 +1,64 @@
+#!/bin/bash
+# test_edit_history_invalid_role_error.sh - Invalid role error
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with a message
+create_chat_file chat.json user "hello"
+
+# Stage the rebase editor output with invalid role
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+pick 0 [invalid] hello
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -eq 0 ]; then
+    echo "FAIL: invalid_role_error - expected non-zero exit code"
+    exit 1
+fi
+
+if ! echo "$OUTPUT" | grep -qi "invalid\|error"; then
+    echo "FAIL: invalid_role_error - expected error about invalid role"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: invalid_role_error"

tests/test_edit_history_multiple_consecutive_edits.sh

@@ -0,0 +1,136 @@
+#!/bin/bash
+# test_edit_history_multiple_consecutive_edits.sh - Multiple consecutive edits to the same message (changes stack correctly)
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with two messages
+create_chat_file chat.json user "hello" assistant "original"
+
+# First session: Edit message 1 to change content from "original" to "first edit"
+# Staged inputs for first drharness invocation:
+#   Invocation 0: rebase editor
+#   Invocation 1: content editor for message 1
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+pick 0 [user] hello
+edit 1 [assistant] original
+EDITEOF
+
+cat > .editor_input_1 << 'EDITEOF'
+# Previous message:
+
+[0] user: hello
+
+Edit message 1 (role: assistant) below:
+
+-- >8 --
+first edit
+EDITEOF
+
+# Staged inputs for second drharness invocation:
+#   Invocation 2: rebase editor (second session)
+#   Invocation 3: content editor for message 1 (second session)
+cat > .editor_input_2 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+pick 0 [user] hello
+edit 1 [assistant] first edit
+EDITEOF
+
+cat > .editor_input_3 << 'EDITEOF'
+# Previous message:
+
+[0] user: hello
+
+Edit message 1 (role: assistant) below:
+
+-- >8 --
+second edit
+EDITEOF
+
+# Run first edit session
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: multiple_consecutive_edits_same_message - first edit expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify first edit was applied
+msg1_content=$(dr_result chat.json 1)
+if [ "$msg1_content" != "first edit" ]; then
+    echo "FAIL: multiple_consecutive_edits_same_message - after first edit, expected 'first edit', got: $msg1_content"
+    exit 1
+fi
+
+# Run second edit session (counter continues from 2)
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: multiple_consecutive_edits_same_message - second edit expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify second edit was applied (changes stack correctly)
+msg1_content=$(dr_result chat.json 1)
+if [ "$msg1_content" != "second edit" ]; then
+    echo "FAIL: multiple_consecutive_edits_same_message - after second edit, expected 'second edit', got: $msg1_content"
+    exit 1
+fi
+
+echo "PASS: multiple_consecutive_edits_same_message"

tests/test_edit_history_multiple_edits.sh

@@ -0,0 +1,98 @@
+#!/bin/bash
+# test_edit_history_multiple_edits.sh - Multiple edits in one session
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with three messages
+create_chat_file chat.json user "msg0" assistant "msg1" user "msg2"
+
+# Stage the rebase editor output (edit messages 0 and 2)
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+edit 0 [user] msg0
+pick 1 [assistant] msg1
+edit 2 [user] msg2
+EDITEOF
+
+# Stage the edit prompt response for message 0
+cat > .editor_input_1 << 'EDITEOF'
+# (no previous message)
+
+Edit message 0 (role: user) below:
+
+-- >8 --
+edited msg0
+EDITEOF
+
+# Stage the edit prompt response for message 2
+cat > .editor_input_2 << 'EDITEOF'
+# Previous message:
+
+[1] assistant: msg1
+
+Edit message 2 (role: user) below:
+
+-- >8 --
+edited msg2
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: multiple_edits - expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify both messages were edited
+msg0_content=$(dr_result chat.json 0)
+if [ "$msg0_content" != "edited msg0" ]; then
+    echo "FAIL: multiple_edits - message 0 not edited correctly"
+    echo "Expected: 'edited msg0', got: '$msg0_content'"
+    exit 1
+fi
+
+msg2_content=$(dr_result chat.json 2)
+if [ "$msg2_content" != "edited msg2" ]; then
+    echo "FAIL: multiple_edits - message 2 not edited correctly"
+    echo "Expected: 'edited msg2', got: '$msg2_content'"
+    exit 1
+fi
+
+echo "PASS: multiple_edits"

tests/test_edit_history_no_edits_unchanged.sh

@@ -0,0 +1,107 @@
+#!/bin/bash
+# test_edit_history_no_edits_unchanged.sh - No edits, history remains the same
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with three messages
+create_chat_file chat.json user "first message" assistant "second message" user "third message"
+
+# Stage the rebase editor output exactly as generated by build_rebase_content
+# This simulates the user doing nothing and just saving the editor
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+pick 0 [user] first message
+pick 1 [assistant] second message
+pick 2 [user] third message
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: no_edits_history_unchanged - expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify history is unchanged - same number of messages
+len=$(history_length chat.json)
+if [ "$len" != "3" ]; then
+    echo "FAIL: no_edits_history_unchanged - expected 3 messages, got $len"
+    exit 1
+fi
+
+# Verify all messages have their original content
+msg0_content=$(dr_result chat.json 0)
+if [ "$msg0_content" != "first message" ]; then
+    echo "FAIL: no_edits_history_unchanged - message 0 content changed"
+    exit 1
+fi
+
+msg1_content=$(dr_result chat.json 1)
+if [ "$msg1_content" != "second message" ]; then
+    echo "FAIL: no_edits_history_unchanged - message 1 content changed"
+    exit 1
+fi
+
+msg2_content=$(dr_result chat.json 2)
+if [ "$msg2_content" != "third message" ]; then
+    echo "FAIL: no_edits_history_unchanged - message 2 content changed"
+    exit 1
+fi
+
+# Verify all roles are unchanged
+msg0_role=$(dr_role chat.json 0)
+if [ "$msg0_role" != "user" ]; then
+    echo "FAIL: no_edits_history_unchanged - message 0 role changed"
+    exit 1
+fi
+
+msg1_role=$(dr_role chat.json 1)
+if [ "$msg1_role" != "assistant" ]; then
+    echo "FAIL: no_edits_history_unchanged - message 1 role changed"
+    exit 1
+fi
+
+msg2_role=$(dr_role chat.json 2)
+if [ "$msg2_role" != "user" ]; then
+    echo "FAIL: no_edits_history_unchanged - message 2 role changed"
+    exit 1
+fi
+
+echo "PASS: no_edits_history_unchanged"

tests/test_edit_history_pick_all_messages.sh

@@ -0,0 +1,79 @@
+#!/bin/bash
+# test_edit_history_pick_all_messages.sh - Test picking all messages (history unchanged)
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with two messages
+create_chat_file chat.json user "hello" assistant "hi there"
+
+# Stage the rebase editor output (pick all)
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+pick 0 [user] hello
+pick 1 [assistant] hi there
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: pick_all_messages - expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify history is unchanged
+len=$(history_length chat.json)
+if [ "$len" != "2" ]; then
+    echo "FAIL: pick_all_messages - expected 2 messages, got $len"
+    exit 1
+fi
+
+msg0_content=$(dr_result chat.json 0)
+if [ "$msg0_content" != "hello" ]; then
+    echo "FAIL: pick_all_messages - message 0 content changed"
+    exit 1
+fi
+
+msg1_content=$(dr_result chat.json 1)
+if [ "$msg1_content" != "hi there" ]; then
+    echo "FAIL: pick_all_messages - message 1 content changed"
+    exit 1
+fi
+
+echo "PASS: pick_all_messages"

tests/test_edit_history_prompt_noop.sh

@@ -0,0 +1,85 @@
+#!/bin/bash
+# test_edit_history_prompt_noop.sh - No edits at edit prompt keeps message unchanged
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with two messages
+create_chat_file chat.json user "hello" assistant "hi there"
+
+# Stage the rebase editor output (edit message 1)
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+pick 0 [user] hello
+edit 1 [assistant] hi there
+EDITEOF
+
+# Stage the edit prompt response (keep same content - no edits)
+cat > .editor_input_1 << 'EDITEOF'
+# Previous message:
+
+[0] user: hello
+
+Edit message 1 (role: assistant) below:
+
+-- >8 --
+hi there
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: edit_prompt_noop - expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify message 1 content is unchanged
+len=$(history_length chat.json)
+if [ "$len" != "2" ]; then
+    echo "FAIL: edit_prompt_noop - expected 2 messages, got $len"
+    exit 1
+fi
+
+msg1_content=$(dr_result chat.json 1)
+if [ "$msg1_content" != "hi there" ]; then
+    echo "FAIL: edit_prompt_noop - message 1 content should be unchanged"
+    exit 1
+fi
+
+echo "PASS: edit_prompt_noop"

tests/test_edit_history_remove_all_content.sh

@@ -0,0 +1,86 @@
+#!/bin/bash
+# test_edit_history_remove_all_content.sh - Editing a message and removing all content (empty content after edit)
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with two messages
+create_chat_file chat.json user "hello" assistant "hi there"
+
+# Stage the rebase editor output (edit message 1)
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+pick 0 [user] hello
+edit 1 [assistant] hi there
+EDITEOF
+
+# Stage the edit prompt response (empty content)
+cat > .editor_input_1 << 'EDITEOF'
+# Previous message:
+
+[0] user: hello
+
+Edit message 1 (role: assistant) below:
+
+-- >8 --
+
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: edit_remove_all_content - expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify history still has 2 messages
+len=$(history_length chat.json)
+if [ "$len" != "2" ]; then
+    echo "FAIL: edit_remove_all_content - expected 2 messages, got $len"
+    exit 1
+fi
+
+# Verify message 1 content is now empty
+msg1_content=$(dr_result chat.json 1)
+if [ "$msg1_content" != "" ]; then
+    echo "FAIL: edit_remove_all_content - message 1 content should be empty, got: $msg1_content"
+    exit 1
+fi
+
+echo "PASS: edit_remove_all_content"

tests/test_edit_history_special_characters.sh

@@ -0,0 +1,86 @@
+#!/bin/bash
+# test_edit_history_special_characters.sh - Editing messages with special characters in content
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with special characters (quotes, dollar sign)
+create_chat_file chat.json user 'hello "quotes" and $dollar' assistant 'backslash\\here'
+
+# Stage the rebase editor output (edit message 1)
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+pick 0 [user] hello "quotes" and $dollar
+edit 1 [assistant] backslash\here
+EDITEOF
+
+# Stage the edit prompt response (with special characters)
+cat > .editor_input_1 << 'EDITEOF'
+# Previous message:
+
+[0] user: hello "quotes" and $dollar
+
+Edit message 1 (role: assistant) below:
+
+-- >8 --
+edited with "quotes" and $dollar signs
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: edit_special_characters - expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify history still has 2 messages
+len=$(history_length chat.json)
+if [ "$len" != "2" ]; then
+    echo "FAIL: edit_special_characters - expected 2 messages, got $len"
+    exit 1
+fi
+
+# Verify message 1 content has the special characters
+msg1_content=$(dr_result chat.json 1)
+if [ "$msg1_content" != 'edited with "quotes" and $dollar signs' ]; then
+    echo "FAIL: edit_special_characters - message 1 content mismatch, got: $msg1_content"
+    exit 1
+fi
+
+echo "PASS: edit_special_characters"

tests/test_edit_history_then_continue.sh

@@ -0,0 +1,114 @@
+#!/bin/bash
+# test_edit_history_then_continue.sh - Edit then continue conversation normally
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with two messages
+create_chat_file chat.json user "hello" assistant "hi there"
+
+# Stage the rebase editor output (edit message 0)
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+edit 0 [user] hello
+pick 1 [assistant] hi there
+EDITEOF
+
+# Stage the edit prompt response
+cat > .editor_input_1 << 'EDITEOF'
+# (no previous message)
+
+Edit message 0 (role: user) below:
+
+-- >8 --
+hello world
+EDITEOF
+
+# Run drharness with -b flag to edit history
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: edit_then_continue - edit phase: expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify the edit was saved
+msg0_content=$(dr_result chat.json 0)
+if [ "$msg0_content" != "hello world" ]; then
+    echo "FAIL: edit_then_continue - edit not saved correctly"
+    exit 1
+fi
+
+# Now run drharness normally (without -b) to continue the conversation
+# We need a new editor input for the normal flow
+# Reset staged inputs for the second run
+cat > .editor_input_0 << 'EDITEOF'
+[assistant] hi there
+
+-- >8 --
+
+how are you?
+EDITEOF
+
+# Reset the invocation counter for the new run
+echo "0" > .editor_invocation_count
+
+# Create mock LLM response for the debug client
+be_done "I'm doing well, thanks for asking!" > mock_responses.txt
+
+# Run drharness normally (uses DebugClient to read from stdin)
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -d -F chat.json < mock_responses.txt
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: edit_then_continue - continue phase: expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify the history now has 4 messages (original 2 + new user message + new assistant response)
+len=$(history_length chat.json)
+if [ "$len" != "4" ]; then
+    echo "FAIL: edit_then_continue - expected 4 messages after continue, got $len"
+    exit 1
+fi
+
+echo "PASS: edit_then_continue"

tests/test_edit_history_unicode_characters.sh

@@ -0,0 +1,80 @@
+#!/bin/bash
+# test_edit_history_unicode_characters.sh - Unicode characters in message content
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a chat file with unicode characters
+create_chat_file chat.json user "こんにちは" assistant "👋 hello 世界"
+
+# Stage the rebase editor output (pick all)
+cat > .editor_input_0 << 'EDITEOF'
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+pick 0 [user] こんにちは
+pick 1 [assistant] 👋 hello 世界
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: unicode_characters - expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify history still has 2 messages
+len=$(history_length chat.json)
+if [ "$len" != "2" ]; then
+    echo "FAIL: unicode_characters - expected 2 messages, got $len"
+    exit 1
+fi
+
+# Verify message content preserved unicode
+msg0_content=$(dr_result chat.json 0)
+if [ "$msg0_content" != "こんにちは" ]; then
+    echo "FAIL: unicode_characters - message 0 unicode not preserved, got: $msg0_content"
+    exit 1
+fi
+
+msg1_content=$(dr_result chat.json 1)
+if [ "$msg1_content" != "👋 hello 世界" ]; then
+    echo "FAIL: unicode_characters - message 1 unicode not preserved, got: $msg1_content"
+    exit 1
+fi
+
+echo "PASS: unicode_characters"

tests/test_edit_history_very_long_content.sh

@@ -0,0 +1,68 @@
+#!/bin/bash
+# test_edit_history_very_long_content.sh - Very long message content in rebase editor
+
+# Helper function to run drharness and capture output to a temp file
+run_drharness() {
+    "$DRHARNESS" "$@" > "$TEST_TEMP_DIR/drharness_output.txt" 2>&1
+    return $?
+}
+
+# Helper to get the output
+get_output() {
+    cat "$TEST_TEMP_DIR/drharness_output.txt"
+}
+
+# Helper to reset editor state for each test
+reset_editor_state() {
+    rm -f .editor_input_* .editor_invocation_count hist.md
+}
+
+reset_editor_state
+
+# Create a very long message (500 characters)
+long_content=$(printf 'x%.0s' {1..500})
+create_chat_file chat.json user "short" assistant "$long_content"
+
+# Stage the rebase editor output (pick all - just verify it handles long content)
+cat > .editor_input_0 << EDITEOF
+# History Rebase Editor
+#
+# Commands:
+#   pick  - keep the message unchanged
+#   edit  - modify the message content and/or role
+#   drop  - remove the message from history
+#
+# Each line format: <command> <index> [<role>] <display_content>
+#   - index: 0-based message index
+#   - role: [user], [assistant], or [tool] (editable)
+#   - display_content: first line of message (display only, not editable here)
+#
+# Lines starting with # are comments and will be ignored.
+# Empty lines are treated as drops.
+#
+-- >8 --
+pick 0 [user] short
+pick 1 [assistant] $long_content
+EDITEOF
+
+# Run drharness with -b flag
+set +e
+VISUAL="$MOCK_EDITOR" run_drharness -b -F chat.json
+EXIT_CODE=$?
+OUTPUT=$(get_output)
+set -e
+
+if [ ${EXIT_CODE} -ne 0 ]; then
+    echo "FAIL: very_long_content - expected exit code 0, got $EXIT_CODE"
+    echo "Output: $OUTPUT"
+    exit 1
+fi
+
+# Verify history still has 2 messages
+len=$(history_length chat.json)
+if [ "$len" != "2" ]; then
+    echo "FAIL: very_long_content - expected 2 messages, got $len"
+    exit 1
+fi
+
+echo "PASS: very_long_content"

tests/test_edit_prompt_first_message.sh

@@ -0,0 +1,45 @@
+#!/bin/bash
+# test_edit_prompt_first_message.sh - Test edit first message (no previous message)
+
+test_name="edit_first_message"
+
+json=$(create_history_json user "hello" assistant "hi there")
+
+OUTPUT=$("${DRHARNESS}" -W "$json" "0" "user")
+
+# Check for "no previous message" text
+if ! echo "$OUTPUT" | grep -q "# (no previous message)"; then
+    echo "FAIL: $test_name - should show no previous message"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Should not have "Previous message:" header
+if echo "$OUTPUT" | grep -q "# Previous message:"; then
+    echo "FAIL: $test_name - should not show previous message header for index 0"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Check for separator
+if ! echo "$OUTPUT" | grep -q -- "-- >8 --"; then
+    echo "FAIL: $test_name - missing separator"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Check for edit prompt
+if ! echo "$OUTPUT" | grep -q "Edit message 0 (role: user) below:"; then
+    echo "FAIL: $test_name - missing edit prompt"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Check for current content
+if ! echo "$OUTPUT" | grep -q "hello"; then
+    echo "FAIL: $test_name - missing current content"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_edit_prompt_middle_message.sh

@@ -0,0 +1,38 @@
+#!/bin/bash
+# test_edit_prompt_middle_message.sh - Test edit middle message shows only immediate previous as context
+
+test_name="edit_middle_message"
+
+json=$(create_history_json user "msg0" assistant "msg1" tool "msg2")
+
+OUTPUT=$("${DRHARNESS}" -W "$json" "2" "tool")
+
+# Should show message 1 as context (immediate previous)
+if ! echo "$OUTPUT" | grep -q "\[1\] assistant: msg1"; then
+    echo "FAIL: $test_name - should show message 1 as context"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Should NOT show message 0 in context
+if echo "$OUTPUT" | grep -q "\[0\]"; then
+    echo "FAIL: $test_name - should not show message 0 in context"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Should show edit prompt for message 2
+if ! echo "$OUTPUT" | grep -q "Edit message 2 (role: tool) below:"; then
+    echo "FAIL: $test_name - should show edit prompt for message 2"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Should show current content (msg2)
+if ! echo "$OUTPUT" | grep -q "msg2"; then
+    echo "FAIL: $test_name - should show current content"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_edit_prompt_multiline_context.sh

@@ -0,0 +1,23 @@
+#!/bin/bash
+# test_edit_prompt_multiline_context.sh - Test multi-line content in context shows full content
+
+test_name="multiline_context"
+
+json=$(create_history_json user $'line one\nline two' assistant "response")
+
+OUTPUT=$("${DRHARNESS}" -W "$json" "1" "assistant")
+
+# Both lines should appear in context
+if ! echo "$OUTPUT" | grep -q "line one"; then
+    echo "FAIL: $test_name - first line of context missing"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+if ! echo "$OUTPUT" | grep -q "line two"; then
+    echo "FAIL: $test_name - second line of context missing"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_edit_prompt_separator_placement.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+# test_edit_prompt_separator_placement.sh - Test separator correctly placed after edit prompt
+
+test_name="separator_placement"
+
+json=$(create_history_json user "first" assistant "second")
+
+OUTPUT=$("${DRHARNESS}" -W "$json" "1" "assistant")
+
+# The separator should appear after the context and edit prompt
+# Check that "Edit message" is followed by the separator
+if ! echo "$OUTPUT" | grep -A2 "Edit message 1" | grep -q -- "-- >8 --"; then
+    echo "FAIL: $test_name - separator not correctly placed after edit prompt"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_edit_prompt_tool_role.sh

@@ -0,0 +1,17 @@
+#!/bin/bash
+# test_edit_prompt_tool_role.sh - Test edit with tool role
+
+test_name="edit_tool_role"
+
+json=$(create_history_json assistant "call tool" tool "result")
+
+OUTPUT=$("${DRHARNESS}" -W "$json" "1" "tool")
+
+# Check for tool role in edit prompt
+if ! echo "$OUTPUT" | grep -q "Edit message 1 (role: tool) below:"; then
+    echo "FAIL: $test_name - should show tool role in edit prompt"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_edit_prompt_with_context.sh

@@ -0,0 +1,46 @@
+#!/bin/bash
+# test_edit_prompt_with_context.sh - Test edit message with previous message context
+
+test_name="edit_with_context"
+
+# Two messages: editing message 1, so message 0 should be shown as context
+json=$(create_history_json user "hello" assistant "hi there")
+
+OUTPUT=$("${DRHARNESS}" -W "$json" "1" "assistant")
+
+# Check for previous message context
+if ! echo "$OUTPUT" | grep -q "# Previous message:"; then
+    echo "FAIL: $test_name - missing previous message header"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Check for index 0 in context
+if ! echo "$OUTPUT" | grep -q "\[0\] user: hello"; then
+    echo "FAIL: $test_name - missing context message with index"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Check for separator
+if ! echo "$OUTPUT" | grep -q -- "-- >8 --"; then
+    echo "FAIL: $test_name - missing separator"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Check for edit prompt
+if ! echo "$OUTPUT" | grep -q "Edit message 1 (role: assistant) below:"; then
+    echo "FAIL: $test_name - missing edit prompt"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Check for current content
+if ! echo "$OUTPUT" | grep -q "hi there"; then
+    echo "FAIL: $test_name - missing current content"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_harness_config.sh

@@ -0,0 +1,61 @@
+#!/bin/bash
+# test_harness_config.sh - Tests for HarnessConfig with base_url and model fields
+
+# Test 1: Verify config is saved with base_url and model in new conversations
+echo "$(be_done "Response")" | "${DRHARNESS}" \
+    -F "${TEST_TEMP_DIR}/chat.json" \
+    -n \
+    -d \
+    -m "test" 2>/dev/null
+
+# Verify config contains base_url and model
+BASE_URL=$(dr_config_base_url "${TEST_TEMP_DIR}/chat.json")
+MODEL=$(dr_config_model "${TEST_TEMP_DIR}/chat.json")
+
+if [ -z "$BASE_URL" ] || [ "$BASE_URL" = "null" ]; then
+    echo "FAIL: Expected config.base_url to be set"
+    exit 1
+fi
+
+if [ -z "$MODEL" ] || [ "$MODEL" = "null" ]; then
+    echo "FAIL: Expected config.model to be set"
+    exit 1
+fi
+
+# Test 2: Verify -M flag sets model in config
+echo "$(be_done "Response")" | "${DRHARNESS}" \
+    -F "${TEST_TEMP_DIR}/chat2.json" \
+    -n \
+    -d \
+    -M "custom-model" \
+    -m "test" 2>/dev/null
+
+MODEL2=$(dr_config_model "${TEST_TEMP_DIR}/chat2.json")
+if [ "$MODEL2" != "custom-model" ]; then
+    echo "FAIL: Expected config.model to be 'custom-model', got '${MODEL2}'"
+    exit 1
+fi
+
+# Test 3: Verify backward compatibility - loading file without base_url and model
+create_state "${TEST_TEMP_DIR}/old_format.json" user "hello"
+
+echo "$(be_done "Response")" | "${DRHARNESS}" \
+    -F "${TEST_TEMP_DIR}/old_format.json" \
+    -d \
+    -m "test" 2>/dev/null
+
+# Verify the saved file now has base_url and model with defaults
+SAVED_BASE_URL=$(dr_config_base_url "${TEST_TEMP_DIR}/old_format.json")
+SAVED_MODEL=$(dr_config_model "${TEST_TEMP_DIR}/old_format.json")
+
+if [ -z "$SAVED_BASE_URL" ] || [ "$SAVED_BASE_URL" = "null" ]; then
+    echo "FAIL: Expected saved config.base_url to be set"
+    exit 1
+fi
+
+if [ -z "$SAVED_MODEL" ] || [ "$SAVED_MODEL" = "null" ]; then
+    echo "FAIL: Expected saved config.model to be set"
+    exit 1
+fi
+
+echo "All harness config tests passed"

tests/test_json_safe_parse.sh

@@ -3,43 +3,64 @@
 
 # Test parse_json_safely with valid JSON
 OUTPUT=$("${DRHARNESS}" -J '{"key": "value"}')
-if [ "${OUTPUT}" != "valid" ]; then
-    echo "FAIL: parse_json_safely - expected 'valid' for valid JSON, got '${OUTPUT}'"
+if ! echo "$OUTPUT" | jq -e '.key == "value"' > /dev/null 2>&1; then
+    echo "FAIL: parse_json_safely - expected valid JSON with key='value', got '${OUTPUT}'"
     exit 1
 fi
 
 # Test parse_json_safely with valid JSON object (empty)
 OUTPUT=$("${DRHARNESS}" -J '{}')
-if [ "${OUTPUT}" != "valid" ]; then
-    echo "FAIL: parse_json_safely - expected 'valid' for empty object, got '${OUTPUT}'"
+if ! echo "$OUTPUT" | jq -e 'type == "object"' > /dev/null 2>&1; then
+    echo "FAIL: parse_json_safely - expected valid empty object, got '${OUTPUT}'"
     exit 1
 fi
 
 # Test parse_json_safely with valid JSON array
 OUTPUT=$("${DRHARNESS}" -J '[1, 2, 3]')
-if [ "${OUTPUT}" != "valid" ]; then
-    echo "FAIL: parse_json_safely - expected 'valid' for array, got '${OUTPUT}'"
+if ! echo "$OUTPUT" | jq -e 'type == "array" and length == 3' > /dev/null 2>&1; then
+    echo "FAIL: parse_json_safely - expected valid array, got '${OUTPUT}'"
     exit 1
 fi
 
 # Test parse_json_safely with invalid JSON (missing quotes)
-OUTPUT=$("${DRHARNESS}" -J '{key: value}')
-if [ "${OUTPUT}" != "discarded" ]; then
-    echo "FAIL: parse_json_safely - expected 'discarded' for invalid JSON, got '${OUTPUT}'"
+set +e
+OUTPUT=$("${DRHARNESS}" -J '{key: value}' 2>&1)
+EXIT_CODE=$?
+set -e
+if [ ${EXIT_CODE} -eq 0 ]; then
+    echo "FAIL: parse_json_safely - expected non-zero exit code for invalid JSON, got 0"
+    exit 1
+fi
+if ! echo "$OUTPUT" | grep -q "error: invalid JSON"; then
+    echo "FAIL: parse_json_safely - expected 'error: invalid JSON' for invalid JSON, got '${OUTPUT}'"
     exit 1
 fi
 
 # Test parse_json_safely with invalid JSON (random text)
-OUTPUT=$("${DRHARNESS}" -J 'not json at all')
-if [ "${OUTPUT}" != "discarded" ]; then
-    echo "FAIL: parse_json_safely - expected 'discarded' for random text, got '${OUTPUT}'"
+set +e
+OUTPUT=$("${DRHARNESS}" -J 'not json at all' 2>&1)
+EXIT_CODE=$?
+set -e
+if [ ${EXIT_CODE} -eq 0 ]; then
+    echo "FAIL: parse_json_safely - expected non-zero exit code for invalid JSON, got 0"
+    exit 1
+fi
+if ! echo "$OUTPUT" | grep -q "error: invalid JSON"; then
+    echo "FAIL: parse_json_safely - expected 'error: invalid JSON' for random text, got '${OUTPUT}'"
     exit 1
 fi
 
 # Test parse_json_safely with empty string
-OUTPUT=$("${DRHARNESS}" -J '')
-if [ "${OUTPUT}" != "discarded" ]; then
-    echo "FAIL: parse_json_safely - expected 'discarded' for empty string, got '${OUTPUT}'"
+set +e
+OUTPUT=$("${DRHARNESS}" -J '' 2>&1)
+EXIT_CODE=$?
+set -e
+if [ ${EXIT_CODE} -eq 0 ]; then
+    echo "FAIL: parse_json_safely - expected non-zero exit code for empty string, got 0"
+    exit 1
+fi
+if ! echo "$OUTPUT" | grep -q "error: invalid JSON"; then
+    echo "FAIL: parse_json_safely - expected 'error: invalid JSON' for empty string, got '${OUTPUT}'"
     exit 1
 fi
 

tests/test_prompt_file.sh

@@ -0,0 +1,48 @@
+#!/bin/bash
+# test_prompt_file.sh - Validates the -P flag reads prompt from file
+
+# Test 1: Basic prompt file reading
+echo "This is my prompt from a file" > prompt.txt
+be_done 'Response received' | "${DRHARNESS}" -F chat.json -n -P prompt.txt -d
+
+# Verify the prompt from the file was stored in the chat history
+STORED_PROMPT=$(dr_result chat.json 0)
+if [ "${STORED_PROMPT}" != "This is my prompt from a file" ]; then
+    echo "FAIL: prompt_file - prompt not read correctly"
+    echo "Expected: 'This is my prompt from a file'"
+    echo "Got: '${STORED_PROMPT}'"
+    exit 1
+fi
+
+# Test 2: Multi-line prompt file
+cat > multiline_prompt.txt << 'PROMPT'
+Line 1
+Line 2
+Line 3
+PROMPT
+
+be_done 'Multi-line response' | "${DRHARNESS}" -F chat2.json -n -P multiline_prompt.txt -d
+
+# Verify multi-line prompt was stored correctly
+STORED_PROMPT=$(dr_result chat2.json 0)
+EXPECTED_MULTILINE="Line 1
+Line 2
+Line 3"
+
+if [ "${STORED_PROMPT}" != "${EXPECTED_MULTILINE}" ]; then
+    echo "FAIL: prompt_file multiline - prompt not read correctly"
+    echo "Expected:"
+    echo "${EXPECTED_MULTILINE}"
+    echo "Got:"
+    echo "${STORED_PROMPT}"
+    exit 1
+fi
+
+# Test 3: Non-existent file should fail
+if "${DRHARNESS}" -F chat3.json -n -P nonexistent.txt -d < /dev/null 2>stderr_output; then
+    echo "FAIL: prompt_file nonexistent - should have failed"
+    cat stderr_output
+    exit 1
+fi
+
+echo "PASS: prompt_file"

tests/test_rebase_apply_all_lines_deleted_keeps.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+# test_rebase_apply_all_lines_deleted_keeps.sh - All lines deleted (only header) keeps all messages
+
+# Empty commands - simulates deleting all message lines
+lines='[]'
+history=$(create_messages_json user "item 0" assistant "item 1")
+
+OUTPUT=$("${DRHARNESS}" -A "$lines" "$history")
+
+# Should have 2 messages (all kept because no commands specified)
+msg_count=$(echo "$OUTPUT" | grep -o '"role"' | wc -l)
+
+if [ "$msg_count" -ne 2 ]; then
+    echo "FAIL: all_lines_deleted_keeps_all - expected 2 messages, got $msg_count"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: all_lines_deleted_keeps_all"

tests/test_rebase_apply_deleting_lines_drops.sh

@@ -0,0 +1,35 @@
+#!/bin/bash
+# test_rebase_apply_deleting_lines_drops.sh - Deleting lines from rebase editor drops those messages (issue 120)
+
+# Simulate editing a 6-item history where items 4 and 5 are deleted
+# Only items 0-3 are mentioned in the commands
+lines=$(create_rebase_lines pick 0 user pick 1 assistant pick 2 tool pick 3 assistant)
+history=$(create_messages_json \
+    user "item 0" assistant "item 1" tool "item 2" assistant "item 3" \
+    user "item 4" assistant "item 5")
+
+OUTPUT=$("${DRHARNESS}" -A "$lines" "$history")
+
+# Should have 4 messages (items 4 and 5 dropped)
+msg_count=$(echo "$OUTPUT" | grep -o '"role"' | wc -l)
+
+if [ "$msg_count" -ne 4 ]; then
+    echo "FAIL: deleting_lines_drops_messages - expected 4 messages, got $msg_count"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Should not contain item 4 or item 5
+if echo "$OUTPUT" | grep -q '"item 4"'; then
+    echo "FAIL: deleting_lines_drops_messages - item 4 should be dropped"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+if echo "$OUTPUT" | grep -q '"item 5"'; then
+    echo "FAIL: deleting_lines_drops_messages - item 5 should be dropped"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: deleting_lines_drops_messages"

tests/test_rebase_apply_drop_removes.sh

@@ -0,0 +1,25 @@
+#!/bin/bash
+# test_rebase_apply_drop_removes.sh - Drop commands remove messages from result
+
+lines=$(create_rebase_lines drop 0 user pick 1 assistant)
+history=$(create_messages_json user "hello" assistant "hi there")
+
+OUTPUT=$("${DRHARNESS}" -A "$lines" "$history")
+
+# Should only have one message (the assistant response)
+msg_count=$(echo "$OUTPUT" | grep -o '"role"' | wc -l)
+
+if [ "$msg_count" -ne 1 ]; then
+    echo "FAIL: drop_removes_message - expected 1 message, got $msg_count"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Should not contain the dropped message content
+if echo "$OUTPUT" | grep -q '"hello"'; then
+    echo "FAIL: drop_removes_message - dropped message should not be in output"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: drop_removes_message"

tests/test_rebase_apply_edit_preserves_fields.sh

@@ -0,0 +1,23 @@
+#!/bin/bash
+# test_rebase_apply_edit_preserves_fields.sh - Non-edited fields are preserved during edits
+
+lines=$(create_rebase_lines edit 0 assistant)
+history=$(create_messages_json user "hello" reasoning "thought process" tool_call_id "call_123")
+
+OUTPUT=$("${DRHARNESS}" -A "$lines" "$history")
+
+# Check reasoning is preserved
+if ! echo "$OUTPUT" | grep -q '"thought process"'; then
+    echo "FAIL: edit_preserves_fields - reasoning should be preserved"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Check tool_call_id is preserved
+if ! echo "$OUTPUT" | grep -q '"call_123"'; then
+    echo "FAIL: edit_preserves_fields - tool_call_id should be preserved"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: edit_preserves_fields"

tests/test_rebase_apply_edit_preserves_tool_calls.sh

@@ -0,0 +1,16 @@
+#!/bin/bash
+# test_rebase_apply_edit_preserves_tool_calls.sh - tool_calls field preserved during edit
+
+lines=$(create_rebase_lines edit 0 assistant)
+history=$(create_messages_json assistant "calling tool" tool_call "call_1" "test" "{}")
+
+OUTPUT=$("${DRHARNESS}" -A "$lines" "$history")
+
+# tool_calls should be preserved
+if ! echo "$OUTPUT" | grep -q '"tool_calls"'; then
+    echo "FAIL: edit_preserves_tool_calls - tool_calls should be preserved during edit"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: edit_preserves_tool_calls"

tests/test_rebase_apply_edit_updates_role.sh

@@ -0,0 +1,23 @@
+#!/bin/bash
+# test_rebase_apply_edit_updates_role.sh - Edit commands update role
+
+lines=$(create_rebase_lines edit 0 assistant)
+history=$(create_messages_json user "hello world")
+
+OUTPUT=$("${DRHARNESS}" -A "$lines" "$history")
+
+# Check role was changed to assistant
+if ! echo "$OUTPUT" | grep -q '"role".*"assistant"'; then
+    echo "FAIL: edit_updates_role - role should be changed to assistant"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Content should still be preserved
+if ! echo "$OUTPUT" | grep -q '"hello world"'; then
+    echo "FAIL: edit_updates_role - content should be preserved during edit"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: edit_updates_role"

tests/test_rebase_apply_empty_lines_preserves.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+# test_rebase_apply_empty_lines_preserves.sh - Empty lines array preserves all messages
+
+lines='[]'
+history=$(create_messages_json user "hello" assistant "hi")
+
+OUTPUT=$("${DRHARNESS}" -A "$lines" "$history")
+
+# Should have 2 messages (all implicitly picked)
+msg_count=$(echo "$OUTPUT" | grep -o '"role"' | wc -l)
+
+if [ "$msg_count" -ne 2 ]; then
+    echo "FAIL: empty_lines_preserves_all - expected 2 messages, got $msg_count"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: empty_lines_preserves_all"

tests/test_rebase_apply_implicit_pick.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+# test_rebase_apply_implicit_pick.sh - Messages not referenced by any command are dropped (implicit drop)
+
+# Only specify command for index 0, index 1 should be implicitly dropped
+lines=$(create_rebase_lines pick 0 user)
+history=$(create_messages_json user "hello" assistant "hi there")
+
+OUTPUT=$("${DRHARNESS}" -A "$lines" "$history")
+
+# Should have 1 message (index 1 dropped)
+msg_count=$(echo "$OUTPUT" | grep -o '"role"' | wc -l)
+
+if [ "$msg_count" -ne 1 ]; then
+    echo "FAIL: implicit_pick - expected 1 message, got $msg_count"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: implicit_pick"

tests/test_rebase_apply_lenient_orphaned_results.sh

@@ -0,0 +1,28 @@
+#!/bin/bash
+# test_rebase_apply_lenient_orphaned_results.sh - Lenient mode - orphaned tool results are allowed
+
+# Drop the tool call message but keep the result (orphaned)
+lines=$(create_rebase_lines drop 0 assistant pick 1 tool)
+history=$(create_messages_json \
+    assistant "" tool_call "call_1" "test" "" \
+    tool "result" tool_call_id "call_1")
+
+OUTPUT=$("${DRHARNESS}" -A "$lines" "$history")
+
+# Should have 1 message (the orphaned tool result)
+msg_count=$(echo "$OUTPUT" | grep -o '"role"' | wc -l)
+
+if [ "$msg_count" -ne 1 ]; then
+    echo "FAIL: lenient_orphaned_results - expected 1 message (orphaned result), got $msg_count"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# The orphaned tool result should be present
+if ! echo "$OUTPUT" | grep -q '"tool"'; then
+    echo "FAIL: lenient_orphaned_results - orphaned tool result should be present"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: lenient_orphaned_results"

tests/test_rebase_apply_mixed_commands.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+# test_rebase_apply_mixed_commands.sh - Mixed commands (pick, drop, edit)
+
+lines=$(create_rebase_lines pick 0 user drop 1 assistant edit 2 tool)
+history=$(create_messages_json user "hello" assistant "hi" tool "result")
+
+OUTPUT=$("${DRHARNESS}" -A "$lines" "$history")
+
+# Should have 2 messages (index 1 dropped)
+msg_count=$(echo "$OUTPUT" | grep -o '"role"' | wc -l)
+
+if [ "$msg_count" -ne 2 ]; then
+    echo "FAIL: mixed_commands - expected 2 messages, got $msg_count"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: mixed_commands"

tests/test_rebase_apply_pick_preserves.sh

@@ -0,0 +1,28 @@
+#!/bin/bash
+# test_rebase_apply_pick_preserves.sh - Pick commands preserve messages unchanged
+
+lines=$(create_rebase_lines pick 0 user)
+history=$(create_messages_json user "hello world")
+
+OUTPUT=$("${DRHARNESS}" -A "$lines" "$history")
+
+if ! echo "$OUTPUT" | grep -q '"hello world"'; then
+    echo "FAIL: pick_preserves_message - content should be preserved"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+if ! echo "$OUTPUT" | grep -q '"role".*"user"'; then
+    echo "FAIL: pick_preserves_message - role should be preserved"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Check we got 1 message back
+if ! echo "$OUTPUT" | grep -q '"role"'; then
+    echo "FAIL: pick_preserves_message - expected output"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: pick_preserves_message"

tests/test_rebase_content_basic_messages.sh

@@ -0,0 +1,30 @@
+#!/bin/bash
+# test_rebase_content_basic_messages.sh - Test basic messages in rebase content
+
+test_name="basic_messages"
+
+json=$(create_messages_json user "hello" assistant "hi there")
+
+OUTPUT=$("${DRHARNESS}" -Z "$json")
+
+# Check header exists
+if ! echo "$OUTPUT" | grep -q "# History Rebase Editor"; then
+    echo "FAIL: $test_name - missing header"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Check pick commands
+if ! echo "$OUTPUT" | grep -q "pick 0 \[user\] hello"; then
+    echo "FAIL: $test_name - missing pick 0"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+if ! echo "$OUTPUT" | grep -q "pick 1 \[assistant\] hi there"; then
+    echo "FAIL: $test_name - missing pick 1"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_content_empty_message.sh

@@ -0,0 +1,16 @@
+#!/bin/bash
+# test_rebase_content_empty_message.sh - Test empty message shows (empty)
+
+test_name="empty_message"
+
+json=$(create_messages_json user "")
+
+OUTPUT=$("${DRHARNESS}" -Z "$json")
+
+if ! echo "$OUTPUT" | grep -q "pick 0 \[user\] (empty)"; then
+    echo "FAIL: $test_name - should show (empty) for empty content"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_content_header_comments.sh

@@ -0,0 +1,28 @@
+#!/bin/bash
+# test_rebase_content_header_comments.sh - Test header explains pick/edit/drop commands
+
+test_name="header_comments"
+
+json=$(create_messages_json user "test")
+
+OUTPUT=$("${DRHARNESS}" -Z "$json")
+
+if ! echo "$OUTPUT" | grep -q "pick  - keep the message unchanged"; then
+    echo "FAIL: $test_name - missing pick explanation"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+if ! echo "$OUTPUT" | grep -q "edit  - modify the message content and/or role"; then
+    echo "FAIL: $test_name - missing edit explanation"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+if ! echo "$OUTPUT" | grep -q "drop  - remove the message from history"; then
+    echo "FAIL: $test_name - missing drop explanation"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_content_message_indices.sh

@@ -0,0 +1,28 @@
+#!/bin/bash
+# test_rebase_content_message_indices.sh - Test each message on its own line with correct index
+
+test_name="message_indices"
+
+json=$(create_messages_json user "msg1" assistant "msg2" tool "msg3")
+
+OUTPUT=$("${DRHARNESS}" -Z "$json")
+
+if ! echo "$OUTPUT" | grep -q "pick 0 \[user\] msg1"; then
+    echo "FAIL: $test_name - missing index 0"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+if ! echo "$OUTPUT" | grep -q "pick 1 \[assistant\] msg2"; then
+    echo "FAIL: $test_name - missing index 1"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+if ! echo "$OUTPUT" | grep -q "pick 2 \[tool\] msg3"; then
+    echo "FAIL: $test_name - missing index 2"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_content_multiline_content.sh

@@ -0,0 +1,25 @@
+#!/bin/bash
+# test_rebase_content_multiline_content.sh - Test multi-line content shows only first line
+
+test_name="multiline_content"
+
+json=$(create_messages_json user "first line
+second line
+third line")
+
+OUTPUT=$("${DRHARNESS}" -Z "$json")
+
+if ! echo "$OUTPUT" | grep -q "pick 0 \[user\] first line"; then
+    echo "FAIL: $test_name - should show only first line"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Make sure it doesn't include the other lines
+if echo "$OUTPUT" | grep -q "second line"; then
+    echo "FAIL: $test_name - should not include second line in pick command"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_content_multiple_tool_calls.sh

@@ -0,0 +1,16 @@
+#!/bin/bash
+# test_rebase_content_multiple_tool_calls.sh - Test multiple tool calls shows first one with ...
+
+test_name="multiple_tool_calls"
+
+json=$(create_messages_json assistant "" tool_call "call_1" "read_file" "" tool_call "call_2" "write_file" "")
+
+OUTPUT=$("${DRHARNESS}" -Z "$json")
+
+if ! echo "$OUTPUT" | grep -q "pick 0 \[assistant\] \[tool_call: read_file \.\.\.\]"; then
+    echo "FAIL: $test_name - should show first tool call with ..."
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_content_scissors_line.sh

@@ -0,0 +1,27 @@
+#!/bin/bash
+# test_rebase_content_scissors_line.sh - Test scissors line separates comments from commands
+
+test_name="scissors_line"
+
+json=$(create_messages_json user "test")
+
+OUTPUT=$("${DRHARNESS}" -Z "$json")
+
+# Check for scissors line
+if ! echo "$OUTPUT" | grep -q -- "-- >8 --"; then
+    echo "FAIL: $test_name - missing scissors line"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Check that scissors line appears before pick commands
+scissors_line=$(echo "$OUTPUT" | grep -n -- "-- >8 --" | head -1 | cut -d: -f1)
+pick_line=$(echo "$OUTPUT" | grep -n "^pick" | head -1 | cut -d: -f1)
+
+if [ "$scissors_line" -ge "$pick_line" ]; then
+    echo "FAIL: $test_name - scissors line should appear before pick commands"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_content_special_characters.sh

@@ -0,0 +1,17 @@
+#!/bin/bash
+# test_rebase_content_special_characters.sh - Test special characters in content
+
+test_name="special_characters"
+
+json=$(create_messages_json user 'hello "quotes" and $dollar' assistant 'backslash\here')
+
+OUTPUT=$("${DRHARNESS}" -Z "$json")
+
+# Should handle quotes and special characters
+if ! echo "$OUTPUT" | grep -q "pick 0 \[user\]"; then
+    echo "FAIL: $test_name - should handle special characters"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_content_tool_call_display.sh

@@ -0,0 +1,16 @@
+#!/bin/bash
+# test_rebase_content_tool_call_display.sh - Test assistant message with tool_calls display
+
+test_name="tool_call_display"
+
+json=$(create_messages_json assistant "" tool_call "call_1" "run_shell" "")
+
+OUTPUT=$("${DRHARNESS}" -Z "$json")
+
+if ! echo "$OUTPUT" | grep -q "pick 0 \[assistant\] \[tool_call: run_shell\]"; then
+    echo "FAIL: $test_name - should show tool_call display"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_content_tool_message.sh

@@ -0,0 +1,16 @@
+#!/bin/bash
+# test_rebase_content_tool_message.sh - Test tool message shows result content
+
+test_name="tool_message"
+
+json=$(create_messages_json tool "tool result content")
+
+OUTPUT=$("${DRHARNESS}" -Z "$json")
+
+if ! echo "$OUTPUT" | grep -q "pick 0 \[tool\] tool result content"; then
+    echo "FAIL: $test_name - should show tool result content"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_content_unicode_characters.sh

@@ -0,0 +1,23 @@
+#!/bin/bash
+# test_rebase_content_unicode_characters.sh - Test unicode characters in content
+
+test_name="unicode_characters"
+
+json=$(create_messages_json user "こんにちは" assistant "👋 hello 世界")
+
+OUTPUT=$("${DRHARNESS}" -Z "$json")
+
+# Should preserve unicode characters
+if ! echo "$OUTPUT" | grep -q "pick 0 \[user\] こんにちは"; then
+    echo "FAIL: $test_name - should preserve unicode"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+if ! echo "$OUTPUT" | grep -q "pick 1 \[assistant\] 👋 hello 世界"; then
+    echo "FAIL: $test_name - should preserve unicode emoji and CJK"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_content_very_long_content.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+# test_rebase_content_very_long_content.sh - Test very long content (truncation test)
+
+test_name="very_long_content"
+
+# Create a very long string (1000 characters)
+long_content=$(printf 'x%.0s' {1..1000})
+json=$(create_messages_json user "$long_content")
+
+OUTPUT=$("${DRHARNESS}" -Z "$json")
+
+# Should handle long content without crashing
+if ! echo "$OUTPUT" | grep -q "pick 0 \[user\]"; then
+    echo "FAIL: $test_name - should handle long content"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_parser_all_commands.sh

@@ -0,0 +1,41 @@
+#!/bin/bash
+# test_rebase_parser_all_commands.sh - Valid rebase file with all three commands
+
+test_name="valid_rebase_all_commands"
+
+content="pick 0 [user] hello world
+edit 1 [assistant] ai response
+drop 2 [tool] tool result"
+
+OUTPUT=$("${DRHARNESS}" -z 3 "$content")
+
+count=$(rebase_line_count "$OUTPUT")
+if [ "$count" -ne 3 ]; then
+    echo "FAIL: $test_name - expected 3 lines, got $count"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+if [ "$(rebase_command_at "$OUTPUT" 0)" != "pick" ]; then
+    echo "FAIL: $test_name - expected pick at 0"
+    exit 1
+fi
+if [ "$(rebase_index_at "$OUTPUT" 0)" != "0" ]; then
+    echo "FAIL: $test_name - expected index 0 at 0"
+    exit 1
+fi
+if [ "$(rebase_role_at "$OUTPUT" 0)" != "user" ]; then
+    echo "FAIL: $test_name - expected role user at 0"
+    exit 1
+fi
+
+if [ "$(rebase_command_at "$OUTPUT" 1)" != "edit" ]; then
+    echo "FAIL: $test_name - expected edit at 1"
+    exit 1
+fi
+if [ "$(rebase_command_at "$OUTPUT" 2)" != "drop" ]; then
+    echo "FAIL: $test_name - expected drop at 2"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_parser_all_valid_roles.sh

@@ -0,0 +1,25 @@
+#!/bin/bash
+# test_rebase_parser_all_valid_roles.sh - All three valid roles accepted
+
+test_name="all_valid_roles"
+
+content="pick 0 [user] hello
+pick 1 [assistant] response
+pick 2 [tool] result"
+
+OUTPUT=$("${DRHARNESS}" -z 3 "$content")
+
+if [ "$(rebase_role_at "$OUTPUT" 0)" != "user" ]; then
+    echo "FAIL: $test_name - expected user role at 0"
+    exit 1
+fi
+if [ "$(rebase_role_at "$OUTPUT" 1)" != "assistant" ]; then
+    echo "FAIL: $test_name - expected assistant role at 1"
+    exit 1
+fi
+if [ "$(rebase_role_at "$OUTPUT" 2)" != "tool" ]; then
+    echo "FAIL: $test_name - expected tool role at 2"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_parser_comments_and_empty_lines_mixed.sh

@@ -0,0 +1,24 @@
+#!/bin/bash
+# test_rebase_parser_comments_and_empty_lines_mixed.sh - Comments and empty lines mixed with commands
+
+test_name="comments_and_empty_lines_mixed"
+
+content="# Header comment
+pick 0 [user] hello
+
+# Middle comment
+edit 1 [assistant] response
+
+drop 2 [tool] result
+# Footer comment"
+
+OUTPUT=$("${DRHARNESS}" -z 3 "$content")
+
+count=$(rebase_line_count "$OUTPUT")
+if [ "$count" -ne 3 ]; then
+    echo "FAIL: $test_name - expected 3 lines, got $count"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_parser_duplicate_indices.sh

@@ -0,0 +1,20 @@
+#!/bin/bash
+# test_rebase_parser_duplicate_indices.sh - Duplicate indices produces error
+
+test_name="duplicate_indices"
+
+content="pick 0 [user] hello
+pick 1 [assistant] response
+drop 0 [tool] duplicate"
+
+set +e
+OUTPUT=$("${DRHARNESS}" -z 3 "$content" 2>&1) || true
+set -e
+
+if ! echo "$OUTPUT" | grep -q "duplicate index 0"; then
+    echo "FAIL: $test_name - expected 'duplicate index 0' error"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_parser_empty_rebase.sh

@@ -0,0 +1,14 @@
+#!/bin/bash
+# test_rebase_parser_empty_rebase.sh - Empty rebase file is valid (no lines)
+
+test_name="empty_rebase"
+
+OUTPUT=$("${DRHARNESS}" -z 3 "")
+
+count=$(rebase_line_count "$OUTPUT")
+if [ "$count" -ne 0 ]; then
+    echo "FAIL: $test_name - expected 0 lines, got $count"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_parser_first_error_returned.sh

@@ -0,0 +1,33 @@
+#!/bin/bash
+# test_rebase_parser_first_error_returned.sh - First error is returned (not all errors aggregated)
+
+test_name="first_error_returned"
+
+content="invalid 0 [user] bad command
+pick 10 [assistant] out of range
+drop 100 [tool] also out of range"
+
+set +e
+OUTPUT=$("${DRHARNESS}" -z 3 "$content" 2>&1) || true
+set -e
+
+if ! echo "$OUTPUT" | grep -q "invalid command: 'invalid'"; then
+    echo "FAIL: $test_name - expected 'invalid command' as first error"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+# Should NOT see the other errors since we exit on first error
+if echo "$OUTPUT" | grep -q "index 10 out of range"; then
+    echo "FAIL: $test_name - should not see second error (exit on first)"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+if echo "$OUTPUT" | grep -q "index 100 out of range"; then
+    echo "FAIL: $test_name - should not see third error (exit on first)"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_parser_index_out_of_range.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+# test_rebase_parser_index_out_of_range.sh - Out of range index produces error
+
+test_name="index_out_of_range"
+
+content="pick 0 [user] hello
+pick 10 [assistant] response"
+
+set +e
+OUTPUT=$("${DRHARNESS}" -z 3 "$content" 2>&1) || true
+set -e
+
+if ! echo "$OUTPUT" | grep -q "index 10 out of range"; then
+    echo "FAIL: $test_name - expected 'index 10 out of range' error"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_parser_invalid_command.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+# test_rebase_parser_invalid_command.sh - Invalid command produces error
+
+test_name="invalid_command"
+
+content="pick 0 [user] hello
+squash 1 [assistant] response"
+
+set +e
+OUTPUT=$("${DRHARNESS}" -z 3 "$content" 2>&1) || true
+set -e
+
+if ! echo "$OUTPUT" | grep -q "invalid command: 'squash'"; then
+    echo "FAIL: $test_name - expected 'invalid command: squash' error"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_parser_invalid_command_digit.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+# test_rebase_parser_invalid_command_digit.sh - Line starting with digit (invalid command) produces error
+
+test_name="invalid_command_digit"
+
+content="0 [user] no command"
+
+set +e
+OUTPUT=$("${DRHARNESS}" -z 3 "$content" 2>&1) || true
+set -e
+
+if ! echo "$OUTPUT" | grep -q "invalid command: '0'"; then
+    echo "FAIL: $test_name - expected 'invalid command: 0' error"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_parser_invalid_index.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+# test_rebase_parser_invalid_index.sh - Invalid index produces error
+
+test_name="invalid_index"
+
+content="pick [user] no valid index"
+
+set +e
+OUTPUT=$("${DRHARNESS}" -z 3 "$content" 2>&1) || true
+set -e
+
+if ! echo "$OUTPUT" | grep -q "invalid index: '\[user\]'"; then
+    echo "FAIL: $test_name - expected 'invalid index' error"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_parser_invalid_role_value.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+# test_rebase_parser_invalid_role_value.sh - Invalid role value produces error
+
+test_name="invalid_role_value"
+
+content="pick 0 [user] hello
+pick 1 [system] response"
+
+set +e
+OUTPUT=$("${DRHARNESS}" -z 3 "$content" 2>&1) || true
+set -e
+
+if ! echo "$OUTPUT" | grep -q "invalid role: '\[system\]'"; then
+    echo "FAIL: $test_name - expected 'invalid role' error"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_parser_malformed_role.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+# test_rebase_parser_malformed_role.sh - Missing role brackets produces error
+
+test_name="malformed_role"
+
+content="pick 0 [user] hello
+pick 1 assistant response"
+
+set +e
+OUTPUT=$("${DRHARNESS}" -z 3 "$content" 2>&1) || true
+set -e
+
+if ! echo "$OUTPUT" | grep -q "malformed role: 'assistant'"; then
+    echo "FAIL: $test_name - expected 'malformed role' error"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_parser_missing_role.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+# test_rebase_parser_missing_role.sh - Missing role produces error
+
+test_name="missing_role"
+
+content="pick 0"
+
+set +e
+OUTPUT=$("${DRHARNESS}" -z 3 "$content" 2>&1) || true
+set -e
+
+if ! echo "$OUTPUT" | grep -q "missing role"; then
+    echo "FAIL: $test_name - expected 'missing role' error"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_parser_only_comments.sh

@@ -0,0 +1,17 @@
+#!/bin/bash
+# test_rebase_parser_only_comments.sh - Only comments produces 0 lines
+
+test_name="only_comments"
+
+content="# comment 1
+# comment 2"
+
+OUTPUT=$("${DRHARNESS}" -z 3 "$content")
+
+count=$(rebase_line_count "$OUTPUT")
+if [ "$count" -ne 0 ]; then
+    echo "FAIL: $test_name - expected 0 lines, got $count"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_parser_whitespace_variations.sh

@@ -0,0 +1,20 @@
+#!/bin/bash
+# test_rebase_parser_whitespace_variations.sh - Pick commands with whitespace variations
+
+test_name="whitespace_variations"
+
+# Test with multiple spaces between fields
+content="pick   0   [user]   hello
+edit  1  [assistant]  response
+drop	2	[tool]	result"
+
+OUTPUT=$("${DRHARNESS}" -z 3 "$content")
+
+count=$(rebase_line_count "$OUTPUT")
+if [ "$count" -ne 3 ]; then
+    echo "FAIL: $test_name - expected 3 lines, got $count"
+    echo "Got: $OUTPUT"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_parser_with_comments.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+# test_rebase_parser_with_comments.sh - Valid rebase with comments ignored
+
+test_name="valid_rebase_with_comments"
+
+content="# This is a comment
+pick 0 [user] hello
+# Another comment
+edit 1 [assistant] response"
+
+OUTPUT=$("${DRHARNESS}" -z 3 "$content")
+
+count=$(rebase_line_count "$OUTPUT")
+if [ "$count" -ne 2 ]; then
+    echo "FAIL: $test_name - expected 2 lines, got $count"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_rebase_parser_with_empty_lines.sh

@@ -0,0 +1,20 @@
+#!/bin/bash
+# test_rebase_parser_with_empty_lines.sh - Valid rebase with empty lines
+
+test_name="valid_rebase_with_empty_lines"
+
+content="pick 0 [user] hello
+
+edit 1 [assistant] response
+
+drop 2 [tool] result"
+
+OUTPUT=$("${DRHARNESS}" -z 3 "$content")
+
+count=$(rebase_line_count "$OUTPUT")
+if [ "$count" -ne 3 ]; then
+    echo "FAIL: $test_name - expected 3 lines, got $count"
+    exit 1
+fi
+
+echo "PASS: $test_name"

tests/test_tool_shell.sh

@@ -41,6 +41,15 @@ timeout 5s "${DRHARNESS}" -F chat3.json -n -m "This is a test" -t run_dangerous_
 
 echo "PASS: tool_shell"
 
+tool_call run_dangerous_shell '{"command":"printf \"test\"; sleep 10 &","timeout_seconds":1}' >> background_task_test.txt
+be_done 'done' >> background_task_test.txt
+
+# If we hit the timeout of the timeout utility then we fail
+timeout 5s "${DRHARNESS}" -F background_task_test.json -n -m "This is a test" -t run_dangerous_shell -d < "background_task_test.txt"
+
+echo "PASS: background_task_test"
+
+
 
 seq 1 1044 > long_file
 tool_call run_dangerous_shell '{"command":"cat long_file"}' >> mock_responses4.txt

tests/test_tool_shell_ctrlc.sh

@@ -0,0 +1,30 @@
+#!/bin/bash
+# test_tool_shell_ctrlc.sh - Test ctrlc (SIGINT) handling in the shell tool
+
+# Test 1: Verify that SIGINT during shell execution results in "[user requested halt of program]" message
+tool_call run_dangerous_shell '{"command":"sleep 100","timeout_seconds":300}' >> mock_ctrlc.txt
+be_done 'Done after interrupt' >> mock_ctrlc.txt
+
+# Run drharness in background, then send SIGINT after a short delay
+"${DRHARNESS}" -F chat_ctrlc.json -n -m "Run shell with ctrlc" -t run_dangerous_shell -d < "mock_ctrlc.txt" > /dev/null 2>&1 &
+DRHARNESS_PID=$!
+
+# Wait a moment for the shell command to start
+sleep 1
+
+# Send SIGINT (Ctrl+C) to the drharness process
+kill -SIGINT $DRHARNESS_PID
+
+# Wait for drharness to complete
+wait $DRHARNESS_PID
+
+# Check that the tool result contains the ctrlc message
+TOOL_RESULT=$(dr_result chat_ctrlc.json 2)
+
+if ! echo "${TOOL_RESULT}" | grep -q "\[user requested halt of program\]"; then
+    echo "FAIL: Expected '[user requested halt of program]' in tool output"
+    echo "Got: ${TOOL_RESULT}"
+    exit 1
+fi
+
+echo "PASS: tool_shell_ctrlc"

tools/agent_tool.cpp

@@ -48,14 +48,14 @@ Tool make_agent_tool(const AgentProfile& profile,
             }
             ToolRegistry sub_registry = std::move(sub_reg_result.value());
 
-            Harness harness(std::make_unique<OllamaClient>(OllamaConfig{
-                .base_url  = base_url,
-                .model     = profile.model,
+            Harness harness(std::make_unique<OllamaClient>(base_url, profile.model, OllamaConfig{
                 .timeout_s = 120
-            }), sub_registry, {
+            }), sub_registry, HarnessConfig{
                 .max_tool_iterations = 16,
                 .verbose             = true,
-                .depth               = depth
+                .depth               = depth,
+                .base_url            = base_url,
+                .model               = profile.model
             });
 
             auto result = harness.chat(task);

tools/edit_tool.cpp

@@ -35,7 +35,7 @@ Tool make_edit_tool() {
             {"properties", {
                 {"filePath", {
                     {"type", "string"},
-                    {"description", "The absolute path to the file to modify"}
+                    {"description", "The path to the file to modify (relative or absolute)"}
                 }},
                 {"oldString", {
                     {"type", "string"},