diff --git a/.beads/beads.jsonl b/.beads/beads.jsonl
index 2b74175d..9b844a4a 100644
--- a/.beads/beads.jsonl
+++ b/.beads/beads.jsonl
@@ -40,7 +40,7 @@
 {"id":"bd-134","title":"Address gosec security warnings (102 issues)","description":"Security linter warnings: file permissions (0755 should be 0750), G304 file inclusion via variable, G204 subprocess launches. Many are false positives but should be reviewed.","design":"Review each gosec warning. Add exclusions for legitimate cases to .golangci.yml. Fix real security issues (overly permissive file modes).","status":"open","priority":2,"issue_type":"task","created_at":"2025-10-25T13:47:10.719134-07:00","updated_at":"2025-10-25T13:47:10.719134-07:00"}
 {"id":"bd-135","title":"Multi-project MCP context switching (GH#145)","description":"Enable MCP server to manage multiple beads projects in a single session with per-request workspace_root parameter.\n\nCurrent bug: set_context(workspace_root) doesn't actually switch sockets - all operations hit first initialized socket.\n\nUse case: Managing tasks across multiple organizations with different permission models (internal, partners, open source).\n\nArchitecture: Connection pool keyed by workspace_root, each maintaining its own daemon socket connection. Request-scoped routing using ContextVar to avoid global state races.\n\nSee GH#145 for full requirements and user context.","design":"✅ APPROVED WITH MODIFICATIONS by architectural review (2025-10-25)\n\nLSP-style model is correct: Single MCP server → per-project daemons → isolated databases.\n\nCRITICAL CHANGES from review:\n1. **Simplify connection pool**: No LRU eviction initially (typical user: 2-5 projects)\n2. **Add asyncio.Lock**: Prevent race conditions in pool access\n3. **Defer health checks**: Only retry on failure, not preemptive pings\n4. **Handle submodules**: Check local .beads BEFORE git toplevel\n5. **Path canonicalization**: realpath + git toplevel with caching\n\nRISKS MITIGATED:\n- Global _client bug: Replace with connection pool keyed by canonical path\n- Race conditions: Add asyncio.Lock for pool mutations\n- Submodule edge case: Check .beads directory first\n- Stale sockets: Retry once on connection failure\n\nEstimated effort: 2.5-3.5 days (simplified from 2.5-4.5 days)\nConfidence: 8/10","acceptance_criteria":"- Multiple projects can be accessed in single MCP session\n- Per-request workspace_root parameter works on all tools\n- No cross-project data leakage\n- Concurrent calls to different projects work correctly\n- Stale sockets auto-reconnect with retry/backoff\n- Integration tests verify isolation across 2+ temp repos\n- set_context() still works as default fallback","notes":"Review doc: docs/bd-135-architectural-review.md\n\nImplementation order:\n1. bd-139 (connection manager) - Foundation with pool + lock\n2. bd-136 (ContextVar routing) - Per-request workspace\n3. bd-138 (require_context) - Validation\n4. bd-140 (tests) - Concurrency + edge cases\n5. bd-141 (docs) - Usage guide\n6. bd-142 (health checks) - DEFERRED to Phase 2","status":"closed","priority":1,"issue_type":"epic","assignee":"amp","created_at":"2025-10-25T13:59:57.231937-07:00","updated_at":"2025-10-25T14:36:02.046142-07:00","closed_at":"2025-10-25T14:36:02.046142-07:00"}
 {"id":"bd-136","title":"Implement request-scoped routing with ContextVar","description":"Add ContextVar-based routing to avoid global state races during concurrent multi-project calls.\n\nApproach:\n- Define current_workspace: ContextVar[str|None] in server.py\n- Add @with_workspace decorator that resolves workspace_root (via _resolve_workspace_root + realpath)\n- Set ContextVar for duration of tool call, reset after\n- Falls back to set_context default (BEADS_WORKING_DIR) if workspace_root not provided\n- beads_mcp.tools.get_client() reads current_workspace from ContextVar\n\nBlocks: bd-135 (connection manager must exist first)","design":"Decorator pattern with ContextVar for request-scoped workspace routing.\n\n@with_workspace decorator:\n- Extract workspace_root parameter from tool call\n- Resolve via _resolve_workspace_root + realpath\n- Set current_workspace ContextVar for request duration\n- Falls back to BEADS_WORKING_DIR if workspace_root not provided\n- Reset ContextVar after tool completes\n\nApplied to all tools in server.py. _get_client() reads current_workspace.\n\n⚠️ CONCURRENCY GOTCHA (from architectural review):\n- ContextVar doesn't propagate to asyncio.create_task() spawned tasks\n- SOLUTION: Keep tool calls synchronous, no background task spawning\n- If background tasks needed: use contextvars.copy_context()\n\nDocument this limitation in bd-141.","notes":"Blocks on bd-139 (connection pool must exist first).\n\nCRITICAL: Do NOT spawn background tasks within tool implementations.\nContextVar propagation to spawned tasks is unreliable.","status":"closed","priority":1,"issue_type":"task","assignee":"amp","created_at":"2025-10-25T14:00:27.895512-07:00","updated_at":"2025-10-25T14:32:36.531658-07:00","closed_at":"2025-10-25T14:32:36.531658-07:00","dependencies":[{"issue_id":"bd-136","depends_on_id":"bd-135","type":"parent-child","created_at":"2025-10-25T14:00:27.896366-07:00","created_by":"daemon"}]}
-{"id":"bd-137","title":"Add health checks and reconnection logic for stale sockets","description":"Handle stale/broken socket connections after daemon restarts or upgrades.\n\nFeatures:\n- ping/health_check method on client\n- Check before use or periodic health check\n- On failure: drop from pool, attempt reconnect with exponential backoff\n- Clear error propagation if reconnect fails after retries\n- Handle version mismatch after daemon upgrade\n- Handle long-idle connections closed by daemon","design":"Add async ping() to DaemonClient. ConnectionManager.get_client() pings before returning cached client. On failure, del from pool and retry connect. Bounded retry (3-5 attempts) with backoff.","status":"in_progress","priority":1,"issue_type":"task","created_at":"2025-10-25T14:00:27.8967-07:00","updated_at":"2025-10-25T17:28:05.384177-07:00"}
+{"id":"bd-137","title":"Add health checks and reconnection logic for stale sockets","description":"Handle stale/broken socket connections after daemon restarts or upgrades.\n\nFeatures:\n- ping/health_check method on client\n- Check before use or periodic health check\n- On failure: drop from pool, attempt reconnect with exponential backoff\n- Clear error propagation if reconnect fails after retries\n- Handle version mismatch after daemon upgrade\n- Handle long-idle connections closed by daemon","design":"Add async ping() to DaemonClient. ConnectionManager.get_client() pings before returning cached client. On failure, del from pool and retry connect. Bounded retry (3-5 attempts) with backoff.","status":"closed","priority":1,"issue_type":"task","created_at":"2025-10-25T14:00:27.8967-07:00","updated_at":"2025-10-25T17:35:42.762836-07:00","closed_at":"2025-10-25T17:35:42.762836-07:00"}
 {"id":"bd-138","title":"Fix require_context to support per-request workspace_root","description":"Update require_context decorator to pass if either:\n- workspace_root was provided on the tool call (via ContextVar), OR\n- BEADS_WORKING_DIR is set (from set_context)\n\nStop using BEADS_DB as router - treat it only as CLI fallback.\n\nEnsures backward compatibility with set_context() while supporting new per-request routing.","status":"closed","priority":2,"issue_type":"task","created_at":"2025-10-25T14:00:27.896646-07:00","updated_at":"2025-10-25T14:32:36.533618-07:00","closed_at":"2025-10-25T14:32:36.533618-07:00","dependencies":[{"issue_id":"bd-138","depends_on_id":"bd-135","type":"parent-child","created_at":"2025-10-25T14:00:27.89804-07:00","created_by":"daemon"}]}
 {"id":"bd-139","title":"Add connection manager for per-project daemon sockets","description":"Create connection pool for managing multiple daemon socket connections in beads_mcp.tools.\n\nSIMPLIFIED APPROACH (per architectural review):\n- Connection pool dict[str, BdClientBase] keyed by canonical path\n- Add asyncio.Lock for thread-safe pool access (CRITICAL)\n- Path canonicalization with @lru_cache for performance\n- NO LRU eviction (start simple, add if monitoring shows memory issues)\n- NO preemptive health checks (only retry on failure)\n\nFeatures:\n- get_client(workspace_root) → returns client for that project\n- Canonicalization: realpath + git toplevel (handles symlinks)\n- Submodule-aware: Check local .beads BEFORE git toplevel\n- Auto-start daemon if socket missing (existing behavior)\n- Validate version on connect (existing behavior)\n\nCRITICAL: Replace global _client singleton with pool to fix set_context() bug.","design":"Implementation in tools.py:\n\n```python\nfrom contextvars import ContextVar\nimport asyncio\nfrom functools import lru_cache\n\ncurrent_workspace: ContextVar[str | None] = ContextVar('workspace', default=None)\n_connection_pool: Dict[str, BdClientBase] = {}\n_pool_lock = asyncio.Lock()  # Prevent race conditions\n\n@lru_cache(maxsize=128)\ndef _canonicalize_path(path: str) -\u003e str:\n    # 1. realpath to resolve symlinks\n    real = os.path.realpath(path)\n    \n    # 2. Check local .beads (submodule edge case)\n    if os.path.exists(os.path.join(real, \".beads\")):\n        return real\n    \n    # 3. Try git toplevel\n    return _resolve_workspace_root(real)\n\nasync def _get_client() -\u003e BdClientBase:\n    workspace = current_workspace.get() or os.environ.get(\"BEADS_WORKING_DIR\")\n    if not workspace:\n        raise BdError(\"No workspace set\")\n    \n    canonical = _canonicalize_path(workspace)\n    \n    async with _pool_lock:  # CRITICAL: prevent concurrent mutations\n        if canonical not in _connection_pool:\n            _connection_pool[canonical] = create_bd_client(\n                prefer_daemon=True,\n                working_dir=canonical\n            )\n        return _connection_pool[canonical]\n```\n\n~200 LOC total (simplified from ~400-500)","status":"closed","priority":1,"issue_type":"task","assignee":"amp","created_at":"2025-10-25T14:00:27.896896-07:00","updated_at":"2025-10-25T14:31:08.910695-07:00","closed_at":"2025-10-25T14:31:08.910695-07:00","dependencies":[{"issue_id":"bd-139","depends_on_id":"bd-135","type":"parent-child","created_at":"2025-10-25T14:00:27.89917-07:00","created_by":"daemon"}]}
 {"id":"bd-14","title":"Auto-flush writes test pollution and session work to git-tracked issues.jsonl","description":"Auto-flush exports ALL issues from DB to issues.jsonl every 5 seconds, including:\n- Test issues (bd-4053 through bd-4059 were version test junk)\n- Issues created during debugging sessions\n- Test pollution from stress tests\n- Temporary diagnostic issues\n\nThis pollutes the git-tracked issues.jsonl with garbage that shouldn't be committed.\n\nExample from today:\n- Git had 49 clean issues\n- Our DB grew to 100+ with test junk and session work\n- Auto-flush wrote all 100+ to issues.jsonl\n- Git status showed modified issues.jsonl with 50+ unwanted issues\n\nImpact:\n- Pollutes git history with test/debug garbage\n- Makes code review difficult (noise in diffs)\n- Can't distinguish real work from session artifacts\n- Other team members pull polluted issues\n\nSolutions to consider:\n1. Disable auto-flush by default (require explicit --enable-auto-flush)\n2. Add .beadsignore to exclude issue ID patterns\n3. Make auto-flush only export 'real' issues (exclude test-*)\n4. Require manual 'bd sync' for git commit\n5. Auto-flush to separate file (.beads/session.jsonl vs issues.jsonl)\n\nRelated: bd-117 (test pollution), isolation_test.go (test DB separation)","status":"closed","priority":1,"issue_type":"bug","created_at":"2025-10-22T00:05:10.788996-07:00","updated_at":"2025-10-24T13:51:54.437366-07:00","closed_at":"2025-10-22T01:05:59.459797-07:00"}