Context Engine MCP Server

# Reactive Review Configuration Summary ## Configuration Locations ### 1. Codex CLI Configuration `C:\Users\preda\.codex\config.toml` ### 2. Gemini Antigravity Configuration `C:\Users\preda\.gemini\antigravity\mcp_config.json` ## Environment Variables Added The following environment variables have been added to the `[mcp_servers.context-engine.env]` section to enable comprehensive reactive review functionality with zombie session detection and plan persistence recovery: ### Master Switch - **REACTIVE_ENABLED="true"** - Master switch for all reactive features ### Phase-Specific Features - **REACTIVE_PARALLEL_EXEC="true"** - Phase 2: Enable parallel execution for faster reviews - **REACTIVE_COMMIT_CACHE="true"** - Phase 1: Enable commit-hash keyed caching for better cache consistency - **REACTIVE_SQLITE_BACKEND="true"** - Phase 3: Enable SQLite backend for plan persistence and recovery - **REACTIVE_GUARDRAILS="true"** - Phase 4: Enable guardrails and validation pipeline ### Tuning Parameters - **REACTIVE_MAX_WORKERS="3"** - Maximum parallel workers for concurrent step execution - **REACTIVE_TOKEN_BUDGET="10000"** - Maximum token budget for a single review session - **REACTIVE_CACHE_TTL="300000"** - Cache TTL in milliseconds (5 minutes) - **REACTIVE_STEP_TIMEOUT="60000"** - Step execution timeout in milliseconds (1 minute) - **REACTIVE_MAX_RETRIES="2"** - Maximum retries for failed steps - **REACTIVE_SESSION_TTL="3600000"** - Session TTL in milliseconds - cleanup after 1 hour - **REACTIVE_MAX_SESSIONS="100"** - Maximum number of sessions to keep in memory - **REACTIVE_EXECUTION_TIMEOUT="600000"** - Execution timeout for zombie session detection (10 minutes) ### Optional Configuration - **REACTIVE_SQLITE_PATH** - Path to SQLite database for plan persistence (commented out, uses default: `~/.context-engine/reactive.db`) ## Key Features Enabled ### 1. Zombie Session Detection With `REACTIVE_EXECUTION_TIMEOUT="600000"` (10 minutes), sessions that remain in executing state without progress will be automatically detected and marked as failed. ### 2. Plan Persistence & Recovery With `REACTIVE_SQLITE_BACKEND="true"`, plans are persisted to disk and can be recovered if they're evicted from memory. This prevents false zombie detection when plans are temporarily unavailable in memory. ### 3. Parallel Execution With `REACTIVE_PARALLEL_EXEC="true"` and `REACTIVE_MAX_WORKERS="3"`, review steps can execute concurrently for faster reviews. ### 4. Commit-Keyed Caching With `REACTIVE_COMMIT_CACHE="true"`, context is cached by commit hash for better consistency during PR reviews. ### 5. Guardrails & Validation With `REACTIVE_GUARDRAILS="true"`, secret scanning, token limits, and validation pipelines are enabled. ## Testing the Configuration To test the zombie session detection and plan recovery: ### For Codex CLI: 1. **Restart Codex CLI** for changes to take effect 2. **Start a reactive review** using the `reactive_review_pr` MCP tool 3. **Check session status** using the `get_review_status` tool 4. **Monitor logs** for plan recovery messages ### For Gemini Antigravity: 1. **Restart Gemini Antigravity** for changes to take effect 2. **Start a reactive review** using the `reactive_review_pr` MCP tool 3. **Check session status** using the `get_review_status` tool 4. **Monitor logs** for plan recovery messages ### Expected Log Messages: - `[ReactiveReviewService] Attempting to recover plan ... from disk` - `[ReactiveReviewService] Successfully recovered plan ... from disk` - `[ReactiveReviewService] Zombie detected: Session ... has plan_id ... but plan not found on disk` ## Related Commits - **96e3126**: `fix: improve zombie session detection with plan persistence recovery` - **7d66456**: `feat: use async plan recovery in MCP tool handlers` ## Test Results - ✅ All 355 tests pass - ✅ 47 reactive review related tests pass - ✅ 3 integration tests for zombie session recovery pass - ✅ Plan recovery from disk verified - ✅ Zombie detection for unrecoverable plans verified - ✅ MCP tool handlers use async plan recovery ## Next Steps 1. **Restart both AI assistants** to apply the new configuration: - Restart Codex CLI - Restart Gemini Antigravity 2. Test reactive review functionality with real PR data 3. Monitor session telemetry and cache hit rates 4. Verify plan persistence and recovery in production scenarios ## Summary Both Codex CLI and Gemini Antigravity configurations have been updated with comprehensive reactive review settings. All features including zombie session detection, plan persistence recovery, parallel execution, and guardrails are now enabled on both platforms.