chore: add CHANGELOG.md and bump version to 0.4.0

- Create CHANGELOG.md documenting releases from 0.1.0 to 0.4.0
- Bump version from 0.2.0 to 0.4.0
- Add changelog requirement note to CLAUDE.md
- Update CLAUDE.md with missing AgentEvent::Retry variant
- Update CLAUDE.md with clemitui test structure
- Fix docs/TOOLS.md system prompt path reference

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: evansenter

CHANGELOG.md

@@ -0,0 +1,61 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [0.4.0] - 2026-01-24
+
+### Added
+- **Reedline REPL**: Line editing, persistent history (10,000 entries), Ctrl-R search
+- **clemitui crate**: Standalone TUI library for ACP-compatible agents
+- **Event bus**: Cross-session coordination via SQLite-backed pub/sub
+- **Plan mode**: Structured planning with tool restrictions and user approval
+- **ACP server**: Agent Client Protocol implementation for subagent spawning
+- **TaskOutput tool**: Retrieve results from background tasks
+- **Comprehensive test suite**: PTY-based terminal tests, plan mode tests, ACP simulation tests
+
+### Changed
+- Improved Ctrl-C handling via `tokio::select!` with biased polling
+- Extracted formatting functions to clemitui for reuse by other agents
+- Consolidated tool formatting into pure functions
+
+### Fixed
+- Made `test_command_safety_classification` more deterministic
+- Fixed newlines in tool output formatting
+
+## [0.3.0] - 2026-01-15
+
+### Added
+- **Task tool**: Spawn subagents for complex operations
+- **Diff highlighting**: Syntax highlighting with Catppuccin Mocha theme
+- **Auto-retry**: Automatic retry on transient API failures
+
+### Changed
+- Simplified streaming architecture to full buffering
+- Consolidated event handling
+
+### Fixed
+- Prevent self-confirmation of dangerous commands
+- Normalize spacing before tools and OUT lines
+
+## [0.2.0] - 2026-01-10
+
+### Added
+- MCP server mode (`--mcp-server`)
+- Tool confirmation flow for destructive commands
+- Background task execution
+
+### Changed
+- Refactored to event-driven architecture
+- Improved tool output formatting
+
+## [0.1.0] - 2026-01-01
+
+### Added
+- Initial release
+- Interactive REPL with streaming output
+- Single prompt mode (`-p "prompt"`)
+- Built-in tools: bash, read, write, edit, glob, grep
+- Tool sandboxing with path validation

CLAUDE.md

@@ -76,15 +76,21 @@ Standalone crate for terminal UI, usable by any ACP-compatible agent:
 ```
 crates/clemitui/
 ├── Cargo.toml
-└── src/
-    ├── lib.rs       # Re-exports
-    ├── format.rs    # Primitive formatting functions (tool output, warnings)
-    ├── logging.rs   # OutputSink trait, log_event functions
-    └── text_buffer.rs # TextBuffer for streaming markdown
+├── src/
+│   ├── lib.rs       # Re-exports
+│   ├── format.rs    # Primitive formatting functions (tool output, warnings)
+│   ├── logging.rs   # OutputSink trait, log_event functions
+│   └── text_buffer.rs # TextBuffer for streaming markdown
+└── tests/
+    ├── common/mod.rs        # Shared test helpers (strip_ansi, RAII guards, CaptureSink)
+    ├── acp_simulation_tests.rs  # 29 tests simulating ACP agent patterns
+    └── e2e_tests.rs         # 19 PTY-based tests for actual terminal output
 ```
 
 **Design**: clemitui takes primitive types (strings, durations, token counts), not genai-rs types. This allows it to work with any ACP agent. clemini's format.rs re-exports these and adds genai-rs-specific wrappers.
 
+Run clemitui tests: `cargo test -p clemitui`
+
 ### Event-Driven Architecture
 
 The agent (`src/agent.rs`) is decoupled from UI via channel-based events:
@@ -105,8 +111,9 @@ run_interaction()                    UI Layer
 - `ToolResult(FunctionExecutionResult)` - Tool completed (uses genai-rs type)
 - `ToolOutput(String)` - Tool output to display (emitted by tools via `ToolEmitter` trait)
 - `Complete { interaction_id, response }` - Interaction finished
-- `ContextWarning { used, limit, percentage }` - Context window >80%
+- `ContextWarning(ContextWarning)` - Context window >80%
 - `Cancelled` - User cancelled
+- `Retry { attempt, max_attempts, delay, error }` - API retry in progress
 
 **`EventHandler` trait** (`src/events.rs`): All UI modes implement this trait:
 - `TerminalEventHandler` (`events.rs`) - REPL and non-interactive modes
@@ -154,9 +161,12 @@ Debugging: `LOUD_WIRE=1` logs all HTTP requests/responses.
 
 ## Documentation
 
+- [CHANGELOG.md](CHANGELOG.md) - Version history and notable changes
 - [docs/TOOLS.md](docs/TOOLS.md) - Tool reference, design philosophy, implementation guide
 - [docs/TEXT_RENDERING.md](docs/TEXT_RENDERING.md) - Output formatting guidelines (colors, truncation, spacing)
 
+**Changelog updates required**: Any user-facing changes (new features, behavior changes, bug fixes) must be documented in CHANGELOG.md before merging.
+
 ## Conventions
 
 - Rust 2024 edition (let chains, etc.)
@@ -196,6 +206,9 @@ If you're unsure whether coverage is sufficient, add more tests. Undertesting ca
 - `semantic_integration_tests.rs` - Multi-turn state, error recovery, code analysis
 - `acp_integration_tests.rs` - ACP subagent spawning and communication
 - `background_tasks_tests.rs` - Background task execution and output retrieval
+- `comprehensive_agent_tests.rs` - Agent interaction patterns, tool chaining, error recovery
+- `plan_mode_tests.rs` - Plan mode entry/exit, tool restrictions, state management
+- `terminal_tests.rs` - PTY-based REPL tests (history, ctrl-c, shell escape, builtins)
 - `event_ordering_tests.rs` - Tool output event ordering (no API key required)
 
 Run locally with: `cargo test --test <name> -- --include-ignored --nocapture`

Cargo.toml

@@ -4,7 +4,7 @@ resolver = "3"
 
 [package]
 name = "clemini"
-version = "0.2.0"
+version = "0.4.0"
 edition = "2024"
 rust-version = "1.88"
 description = "A Gemini-powered coding CLI, built with genai-rs"

docs/TOOLS.md

@@ -105,7 +105,7 @@ Every tool needs tests covering:
 
 5. **Register in `CleminiToolService`** in `src/tools/mod.rs`
 
-6. **Update the System Instruction** in `src/main.rs` SYSTEM_PROMPT
+6. **Update the System Instruction** in `src/system_prompt.md`
 
 7. **Run quality gates**: `make clippy && make fmt && make test`