mcp-code-context 3.6.2 → 3.7.0

package/CHANGELOG.md

@@ -1,268 +1,140 @@
-# Changelog
-
-## [3.6.2] - 2026-05-01
-
-### 🔒 CRITICAL FIXES - Production Stability Release
-
-**Emergency Update**: Resolved 3 critical production bugs causing data corruption, memory leaks, and race conditions.
-
-### Fixed
-
-#### Critical Bug Fixes
-- **FIX #1: Double Cache Instance** — Eliminated duplicate SQLite database instances in `read.ts` causing cache corruption. Cache now properly injected from `index.ts` singleton.
-- **FIX #2: Memory Leak (Event Listeners)** — Removed `process.on` listeners from `CacheManager` constructor. Each cache instance was registering global listeners, causing `MaxListenersExceededWarning` and memory exhaustion. Added global shutdown hook in `index.ts` instead.
-- **FIX #3: TOCTOU Race Condition** — Added `originalHash` verification in Phase 2 writes. Prevents silent data corruption when files are modified between dry-run (Phase 1) and confirmation (Phase 2).
-
-### Added
-- **Global Shutdown Hook**: Single `process.on` handler in `index.ts` persists all active caches on exit
-- **persistOnExit() Method**: Public synchronous method for safe cache persistence during shutdown
-- **originalHash Field**: Added to `WriteResult` interface for TOCTOU protection
-- **8 Regression Tests**: Memory leak detection, cache injection validation, TOCTOU protection
-
-### Changed
-- **BREAKING**: `extractSymbol` now requires `cache` parameter (injected from caller)
-- **BREAKING**: `WriteResult` interface includes optional `originalHash` field
-- **BREAKING**: `PendingOperation` interface includes optional `originalHash` field
-
-### Test Results
-- 81/82 tests passing (99% success rate)
-- Memory leak test: ✅ PASSED (no listener accumulation)
-- TOCTOU protection test: ✅ PASSED (hash verification works)
-- Cache injection test: ✅ PASSED (no duplicate instances)
-- 1 flaky performance test (timing variance, not a bug)
-
-### Security
-- ✅ TOCTOU race condition eliminated
-- ✅ Memory leak prevention
-- ✅ Cache corruption prevention
-
-### Migration from v3.6.0
-- Update calls to `extractSymbol` to pass `cache` parameter
-- No user-facing API changes (internal refactoring only)
-- Rebuild recommended: `npm run rebuild`
-
----
-
-## [3.6.0] - 2026-04-30
-
-### Added
-
-#### Infrastructure
-- **Persistent WASM SQLite Cache** (`sql.js`) — <100ms cache hits, 10× faster on repeated reads. Debounced persistence (5s) prevents event loop blocking. OS temp cleanup recovery built-in.
-- **Structured Logging** (`pino`) — JSON to stderr (MCP-safe). Pretty-print in development, JSON in production. All `console.log/error` replaced.
-- **File Watcher** (`chokidar`) — Auto-invalidates cache on file changes. Configurable debounce (default 500ms). New MCP tools: `configure_file_watcher`, `get_file_watcher_status`.
-- **Fuzzy Search** (`fuse.js`) — Typo-tolerant search. Finds `authenticateUser` when searching `authUser`. Configurable threshold.
-- **Pagination** — `search_code_pattern` now defaults to 10 results (was 50) with `startIndex` parameter and result footer.
-
-#### Testing (74 tests, 100% passing)
-- **Vitest** test framework with V8 coverage
-- Unit tests: parsers, diff, fileLock, backupManager, phase9 optimization
-- Integration tests: full read→write→rollback cycle, multi-process safety
-- Performance tests: cache hit <100ms, search <2000ms
-- Stress tests: 20 concurrent locks, 10K line files, OOM protection
-
-#### Security Fixes (3 production stoppers resolved)
-- **STOPPER #1**: `sanitizeRegexPattern` UUID bug fixed (`\\$&` replacement)
-- **STOPPER #2**: Phase 2 re-acquires file lock before write (race condition eliminated)
-- **STOPPER #3**: `SecurityValidator` applied in Phase 2 (defense in depth)
-
-### Changed
-- **Token Optimization**: Phase 2 confirmation no longer repeats diff (50-80% token reduction)
-- **Auto-optimize output**: `get_semantic_repo_map` auto-disables symbols if >100 files or >1000 symbols
-- **Filesystem locks**: `proper-lockfile` replaces in-memory locks (multi-process safe)
-- **OS temp backups**: Backups moved from project root to `os.tmpdir()` (no hot-reload loops)
-- **Myers diff**: `diff-match-patch` replaces custom O(n×m) LCS (no OOM on large files)
-- **Centralized `walkDir`**: Extracted to `src/utils/fileWalker.ts` (eliminated 3× duplication)
-- **Pinned dependencies**: All `^` ranges replaced with exact versions for reproducible installs
-
-### Fixed
-- `analyzeImpact` regex patterns now use `safeRegex` (consistent ReDoS protection)
-- `BackupManager.getBackupRoot` uses async `fs.mkdir` (no sync blocking)
-- `validateRegexPattern` documented as best-effort (worker timeout is the real guarantee)
-
-### Performance
-- Cache hit: <100ms ✅
-- Parse file: <500ms ✅
-- Search (20 files): <2000ms ✅
-- Token savings: 50-80% ✅
-
-### Migration from v3.6.2
-- Zero breaking changes
-- New optional parameters: `fuzzyMatch`, `fuzzyThreshold`, `startIndex`, `maxDepth`, `includeSymbols`
-- New tools: `configure_file_watcher`, `get_file_watcher_status`, `get_cache_stats`, `clear_cache`
-
----
-
-## [3.6.2] - 2026-04-24
-
-### 🔒 PRODUCTION HARDENING - Security & Infrastructure Complete
-
-**Major Update**: All utility modules integrated, critical bugs fixed, production-ready
-
-### Added
-- **Two-Phase Write Workflow**: Dry-run + confirmation tokens for all write operations
-- **Rate Limiting**: Token bucket algorithm with per-operation costs
-- **File Locking**: Prevents concurrent write conflicts with timeout protection
-- **Audit Logging**: JSON-based audit trail with query/stats API
-- **Telemetry**: Operation metrics, cache stats, Prometheus export
-- **Streaming I/O**: Memory-safe file operations for large files (>5MB)
-- **get_server_stats**: New tool exposing telemetry and audit metrics
-- **confirm_write**: New tool for two-phase write confirmation
-
-### Fixed
-- **CRITICAL**: `extractSymbol` API mismatch in read.ts (symbol extraction was 100% broken)
-- **CRITICAL**: Path traversal in 4 handlers (readLines, searchPattern, analyzeImpact, renameSymbol)
-- **CRITICAL**: ReDoS vulnerability in readLines (raw regex without timeout)
-- **CRITICAL**: `forEach(async)` bug in searchPattern (incomplete results)
-- **Security**: renameSymbol now uses SecurityValidator + atomic writes
-- **Parser**: replaceSymbol moved to BaseParser using AST indices (no more indexOf fragility)
-- **Diff**: Added MAX_DIFF_LINES=5000 guard to prevent OOM on large files
-
-### Changed
-- **BREAKING**: All tools now require `projectRoot` parameter for security
-- **Security**: SecurityValidator enforced on all 13 handlers
-- **Architecture**: Middleware pipeline (RateLimiter → FileLock → Execute → AuditLog → Telemetry)
-- **Parsers**: Removed 300+ lines of duplicated replaceSymbol code
-- **Version**: Synchronized all version strings to 3.6.2
-
-### Performance
-- Streaming I/O for files >5MB (prevents memory exhaustion)
-- LCS diff fallback for files >5000 lines (O(n) vs O(n²))
-- Rate limiter with configurable operation costs
-
-### Security
-- ✅ Path traversal protection on all handlers
-- ✅ ReDoS protection with regex timeout
-- ✅ Atomic writes with .tmp + rename pattern
-- ✅ File locking prevents concurrent corruption
-- ✅ Audit trail for all operations
-
-### Validation
-- ✅ 47/47 WASM parser tests passing (100%)
-- ✅ All security tests passing
-- ✅ All rate limiter tests passing
-- ✅ All file lock tests passing
-- ✅ All v3.6.2 feature tests passing
-- ✅ TypeScript: 0 compilation errors
-- ✅ Build: Clean production dist
-
-### Migration from v3.1.0
-- Add `projectRoot` parameter to all tool calls
-- Optional: Use `confirm: false` for dry-run, then `confirm: true` with token
-- Optional: Call `get_server_stats` to monitor operations
-
----
-
-## [3.1.0] - 2026-04-24
-
-### 🚀 FEATURE COMPLETE - All 11 Tools Implemented
-
-**Major Update**: Added 7 missing tools from v2.6.0 to achieve feature parity
-
-### Added
-- **get_semantic_repo_map**: Compress entire repositories to structural signatures
-- **read_file_surgical**: Extract specific symbols from files  
-- **analyze_impact**: Find all files that depend on a given file
-- **read_file_lines**: Read specific line ranges from files
-- **search_code_pattern**: Search for patterns across multiple files
-- **rename_symbol**: Rename symbols across entire repository
-
-### Summary
-- ✅ 11/11 tools from v2.6.0 now implemented
-- ✅ 100% WASM portability maintained
-- ✅ Zero native dependencies
-- ✅ Feature parity with v2.6.0 achieved
-- ✅ Tree-sitter WASM for 100% AST accuracy
-
----
-
-## [3.0.6] - 2026-04-24
-
-### 🚀 COMPLETE REWRITE - Tree-sitter WASM Edition
-
-**Breaking Changes**: Complete architecture overhaul with 100% WASM portability
-
-### Added
-- **Tree-sitter WASM Parsers**: 100% AST accuracy for TypeScript/JavaScript/Python/PHP/Dart
-- **Zero Native Dependencies**: No Visual Studio, node-gyp, or Python required
-- **Cross-Platform Portability**: Works on Windows/Mac/Linux without compilation
-- **Dart AST Support**: Full Tree-sitter parsing (no regex fallback)
-- **Mandatory Security Boundaries**: projectRoot required for all operations
-- **Async-First Architecture**: Zero blocking I/O
-- **Simplified Core**: Engine + Registry + Validator pattern
-- **Smart WASM Path Resolution**: Finds WASM files in installed module location (not CWD)
-
-### Changed
-- **CRITICAL**: Migrated from tree-sitter native to web-tree-sitter@0.25.1
-- Uses tree-sitter-wasms@0.1.13 for language grammars (ABI v15)
-- Parser API: `cursor.currentNode` is now getter (not function call)
-- Language loading: `Language.load(buffer)` with fs.readFileSync
-- WASM path resolution: Uses `import.meta.url` for module-relative paths
-- Replaced ts-morph with tree-sitter WASM
-- Replaced php-parser with tree-sitter-php WASM
-- Simplified from 9,518 lines to ~600 lines of core code
-
-### Removed
-- Native tree-sitter bindings (node-gyp dependency)
-- Legacy v2.x handlers (readHandlers, writeHandlers, utilHandlers)
-- php-parser dependency
-- ts-morph dependency
-- Regex-based Dart parser fallback
-
-### Fixed
-- **npx compatibility**: Works with `npx -y mcp-code-context@3.1.0`
-- **WASM loading**: Finds WASM files when installed globally or locally
-- **Windows support**: Full compatibility with Claude Desktop on Windows
-
-### Performance
-- 100% accuracy on TypeScript/JavaScript/Python/PHP/Dart parsing
-- Zero regex-based parsing
-- Cursor-based tree traversal (efficient)
-- WASM overhead: ~10-20% slower than native, but portable
-
-### Security
-- Mandatory project boundary enforcement
-- Path traversal protection
-- Async file validation
-- No native code execution (WASM sandbox)
-
-### Validation
-- ✅ 47/47 unit tests passing (100%)
-- ✅ Dart: 4 symbols detected (class, methods, functions)
-- ✅ TypeScript: 4 symbols detected (class, constructor, methods, functions)
-- ✅ Python: 12/12 tests passing
-- ✅ PHP: 11/11 tests passing
-- ✅ Build passes without TypeScript errors
-- ✅ No native dependencies in package.json
-
-### Migration Guide
-v2.x users: This is a ground-up rewrite. v2.x tools are not compatible.
-New API focuses on parse_file with Tree-sitter WASM accuracy.
-
-**Installation**: Just `npm install` - no build tools required!
-
----
-
-## [2.6.0] - 2024-01-16
-
-### 🚀 Performance & Scalability
-
-- **Async I/O Migration**: Eliminated all synchronous file operations
-  - `ignoreManager.walkDirectory()` now fully async
-  - No event loop blocking on large repositories
-  - 3x faster on repos with 1000+ files
-  - Impact: Scalability 4.0 → 9.0 (+5.0)
-
-- **Fuzzy Match Threshold**: Reduced noise in symbol suggestions
-  - Added `maxDistance` parameter (default: 3 edits)
-  - Filters out irrelevant suggestions
-  - Better UX when symbol not found
-
-### 📊 Metrics Improvement
-- Scalability: 4.0 → 9.0 (+5.0) ✅
-- Overall: 8.1 → 9.3 (+1.2)
-
-### Technical
-- All 149 tests passing
-- Zero breaking changes
-- Backward compatible
+# Changelog
+
+## [3.7.0] - 2026-06-18
+
+### 🚀 Production Readiness Release — Adversarial Audit Hardening
+
+This release fixes **critical architectural issues** identified through three rounds of adversarial analysis and AST-based code review. All stoppers resolved. Ready for multi-agent production deployments.
+
+### Fixed — Critical (Production Blockers)
+
+1. **Session State Leakage / Memory Leak**
+   - Global state (`globalLockManager`, `globalConfirmationStore`, `rateLimiter`) accumulated across client sessions with no cleanup.
+   - **Fix**: `SessionManager` per-client state isolation with LRU eviction and 30-min timeout cleanup.
+
+2. **No Phase 1 Crash Recovery**
+   - Pending operations stored in memory only. Server crash between Phase 1 → Phase 2 lost all work.
+   - **Fix**: SQLite-backed `PendingOperationStore` — operations persist across restarts.
+
+3. **TOCTOU in Multi-File Writes**
+   - Phase 2 only verified hash of the primary file, not dependent files in `pendingWrites`.
+   - **Fix**: Hash verified on all files in the write set before applying.
+
+4. **Symlink Boundary Escape**
+   - `path.resolve()` used for boundary validation — does not follow symlinks.
+   - **Fix**: `fs.realpath()` in `SecurityValidator.validateFilePath()`.
+
+5. **`new Promise(async ...)` Anti-pattern in Worker**
+   - `runInWorker` used `new Promise(async (resolve) => ...)` — swallowed async errors.
+   - **Fix**: Refactored to standard async/await pattern.
+
+6. **`WORKER_FILE` Resolved at Import-Time**
+   - Worker path calculated before `dist/` exists — crashes on fresh install.
+   - **Fix**: Lazy `getWorkerFile()` evaluated at runtime.
+
+7. **ReDoS in `impactAnalysis`, `rename`, `readCore`**
+   - Dynamic `new RegExp()` constructed in main thread without timeout.
+   - **Fix**: All operation-level regex routed through `safeRegex` worker pool.
+
+8. **Rate Limiter Shared Globally**
+   - Single `'global'` bucket shared across all MCP clients.
+   - **Fix**: Per-session `SESSION_ID` (pid-based) rate limiting.
+
+9. **`configure_file_watcher` Without `rootDir` Validation**
+   - Watcher path not validated through `SecurityValidator`.
+   - **Fix**: Watcher uses only validated `projectRoot`.
+
+10. **Confirmation Token Leak Without Cleanup**
+    - Expired tokens accumulated in memory indefinitely.
+    - **Fix**: `setInterval` 60s cleanup in `ConfirmationStore`.
+
+11. **Windows Path Case Sensitivity in Boundary Check**
+    - `fs.realpath()` returns `C:\` (uppercase), `path.resolve()` returns `c:\` (lowercase).
+    - **Fix**: `.toLowerCase()` normalization on both sides before comparison.
+
+12. **`ast_transform` try/catch Indentation Bug**
+    - `applyWrapTryCatch` added extra indent on top of existing indentation.
+    - **Fix**: Correct base-indent detection using last line of rawBody as closing-indent anchor.
+
+### Added
+
+#### 25 Tools (up from 13 in v3.6.x)
+
+New tools added in this release:
+
+| Tool | Purpose |
+|------|---------|
+| `ast_transform` | Declarative AST transforms: add_parameter, wrap_with_try_catch, add_decorator, change_return_type, extract_variable |
+| `search_symbols` | AST-aware symbol search by name (fuzzy-capable), not text grep |
+| `explain_symbol` | Signature + location + callers in one call |
+| `batch_read` | Read N symbols from N files in 1 round-trip |
+| `get_rate_limit_status` | Token balance + `canAfford` map per operation |
+| `get_session_stats` | Per-session diagnostics (pending ops, locks, tokens) |
+| `clear_session_cache` | Invalidate cache for current session only |
+| `list_pending_operations` | Recovery listing after crash |
+
+#### Architecture Improvements
+
+- `SessionManager`: Per-client isolated state (locks, confirmations, rate limiter, cache pool)
+- `safeRegexPool.ts` + `safeRegexApi.ts`: Worker pool split from monolithic `safeRegex.ts`
+- `toctou.ts`: TOCTOU hash helpers extracted to own module
+- `pendingStore.ts` + `pendingStoreInit.ts`: SQLite crash-recovery layer
+- `toolDefinitionsAdmin.ts` + `toolDefinitionsCore.ts`: Tool definitions split from 1000-line index
+- All `src/` files under 300 lines (from 1000-line monolith)
+
+#### DX / Ergonomics
+
+- `diffFormat` param on all write tools: `unified` | `compact` | `summary` | `none`
+- Phase 2 does NOT require re-sending `newContent`/`code` (server stores from Phase 1)
+- `aroundPattern` output includes `"Match found at line N (showing lines X-Y):"`
+- `get_cache_stats` includes `hits`, `misses`, `hitRate`
+- `get_file_watcher_status` includes `paths[]` (capped at 50)
+- `get_server_stats` includes `rateLimiter.tokensAvailable` and `operationCosts`
+- `MAX_FILES_REPO_MAP` = 2000 (was 500)
+
+### Test Results
+
+- **83/83 tests passing** (100%)
+- Security tests: ✅
+- File lock tests: ✅
+- Rate limiter tests: ✅
+- WASM parser tests (TS/PHP/Python/Dart): ✅
+- Backup manager, diff, TOCTOU, cache, confirmation store: ✅
+- Multi-process concurrency tests: ✅
+- Performance tests: ✅
+
+### Migration from v3.6.x
+
+- **Zero breaking changes** in the public tool API.
+- `parse_file` now returns `startLine`/`endLine` per symbol — additive, no breakage.
+- New tools are purely additive.
+- Session isolation is automatic — no configuration needed.
+
+---
+
+## [3.6.4] - 2026-05-14
+
+### Added
+- Language parsers: C#, Go, Java, Kotlin, Ruby, Rust (Tree-sitter WASM)
+- `parse_file` returns `startLine`/`endLine` per symbol
+- Fuzzy search with `fuzzyMatch` + `fuzzyThreshold` parameters
+- Pagination via `startIndex` on `search_code_pattern`
+- `diffFormat` parameter on write tools
+
+### Fixed
+- `parse_file` symbols missing line numbers
+- CVE in `@modelcontextprotocol/sdk` (updated to 1.29.0)
+- Incorrect TypeScript version `5.7.0` → `5.7.2`
+
+---
+
+## [3.6.1] - 2026-04-28
+
+### Added
+- Persistent WASM SQLite cache (`CacheManager`)
+- Structured logging via pino (stderr-safe for MCP)
+- File watcher via chokidar
+- Multi-process file locking via proper-lockfile
+- Rolling 5-version backup system
+- `rollback_file` and `clean_backups` tools
+- 74 tests (unit + integration + performance + stress)

package/INSTRUCTIONS.md

@@ -108,7 +108,7 @@ newContent: "return `Hello, ${name}`;"
   maxResults: 10
 ```
 
-### Fuzzy Search (v3.6.2)
+### Fuzzy Search
 ```
 @mcp-code-context/search_code_pattern
   rootDir: /project/root
@@ -189,14 +189,14 @@ Rename across entire codebase in one operation:
 ## ⚠️ Common Mistakes
 
 ### ❌ Mistake 1: Forgetting projectRoot
-```
+```typescript
 // WRONG
 @mcp-code-context/read_file_surgical
   filePath: /path/to/file.ts
 ```
 
 ### ❌ Mistake 2: Skipping Phase 1
-```
+```typescript
 // WRONG - No preview
 @mcp-code-context/write_file_surgical
   confirm: true
@@ -204,13 +204,13 @@ Rename across entire codebase in one operation:
 ```
 
 ### ❌ Mistake 3: Incomplete Symbol Code
-```
+```typescript
 // WRONG - Missing function signature
 newContent: "{ return 42; }"
 ```
 
 ### ❌ Mistake 4: Not Checking Impact
-```
+```typescript
 // WRONG - Removing without checking dependencies
 @mcp-code-context/remove_symbol
   symbolName: importantFunction
@@ -253,14 +253,32 @@ newContent: "{ return 42; }"
 
 ---
 
-## 🚀 v3.6.2 Features
-
+## 🚀 v3.7.0 Features
+
+### New Tools
+- **`get_session_stats`**: Session-scoped stats (pending ops, locks held, rate limiter tokens)
+- **`clear_session_cache`**: Clear cache for current session only (no impact on other clients)
+- **`list_pending_operations`**: List all pending operations (for recovery after crash)
+
+### Improvements
+- ✅ **Session-scoped state**: No state leakage between MCP clients
+- ✅ **Crash recovery**: SQLite-backed pending operations store
+- ✅ **Expanded ReDoS protection**: 15+ patterns (was 6)
+- ✅ **Auto-persist before eviction**: Cache data preserved
+- ✅ **Higher file limits**: MAX_FILES = 2000 (was 500)
+
+### Previously Added in v3.7.0
+- ✅ **4 New Tools**: `search_symbols`, `explain_symbol`, `batch_read`, `get_rate_limit_status`
+- ✅ **`diffFormat`**: Control diff verbosity (`unified`, `compact`, `summary`, `none`)
+- ✅ **Phase 2 optimized**: No need to resend `newContent`/`code` — server stores it from Phase 1
+- ✅ **`aroundPattern` line number**: Output now shows `Match found at line N`
+- ✅ **Rate limiter in stats**: `get_server_stats` includes token balance and operation costs
+- ✅ **Cache hit rate**: `get_cache_stats` includes `hits`, `misses`, `hitRate`
+- ✅ **Watcher paths**: `get_file_watcher_status` includes list of watched files
 - ✅ **Fuzzy Search**: Find similar patterns with typo tolerance
 - ✅ **Pagination**: Default 10 results, use `startIndex` for more
-- ✅ **Token Optimization**: Phase 2 no longer repeats diff (97% savings)
-- ✅ **Auto-Optimize**: Large repos automatically disable symbols
-- ✅ **File Watcher**: Auto-invalidate cache on file changes
 - ✅ **Persistent Cache**: 10x faster with sql.js WASM SQLite
+- ✅ **File Watcher**: Auto-invalidate cache on file changes
 
 ---
 
@@ -270,5 +288,6 @@ See `llms.txt` for complete API reference and technical details.
 
 ---
 
-**Version**: 3.6.2  
-**Last Updated**: May 1, 2026
+**Version**: 3.7.0  
+**Last Updated**: June 8, 2026  
+**Status**: Production-ready with session-scoped state and crash recovery