MCP tool hooks_route_enhanced returns 'spawnSync /bin/sh ETIMEDOUT' while CLI 'ruvector hooks route-enhanced' runs in <200ms

[ronmikailov]

Summary

The mcp__ruvector__hooks_route_enhanced MCP tool always returns {"success": false, "error": "spawnSync /bin/sh ETIMEDOUT"} for any input. The equivalent CLI subcommand ruvector hooks route-enhanced "<task>" succeeds in ~182ms, so the regression is inside the MCP server's shell-out path, not the underlying routing code.

Environment

• ruvector version: 0.2.25 (latest on npm at time of report)
• Node: v20.19.6
• OS: Linux 6.6.87.2-microsoft-standard-WSL2 (Ubuntu, WSL2)
• MCP client: Claude Code (claude-flow integration)
• Server config: .mcp.json → ruvector mcp start (default, no env overrides)

Reproduction

1. Start the ruvector MCP server (ruvector mcp start) and connect from any MCP client.
2. Call the hooks_route_enhanced tool with any task:

   { "task": "test routing" }

3. Observe response.

Actual

{
  "success": false,
  "error": "spawnSync /bin/sh ETIMEDOUT"
}

The error is reproducible on every call, including immediately after server start (so it is not a warm-up / first-call issue).

Expected

A routing decision similar to what the CLI returns:

$ time ruvector hooks route-enhanced "test routing"
{"success":true,"agent":"coder","confidence":0.5,"reason":"default mapping","signals":[],"coverageWeight":null,"complexity":null}

real    0m0.182s

Comparison: same input through CLI vs MCP

Path	Result	Time
ruvector hooks route-enhanced "test" (CLI)	success, returns routing JSON	~180ms
mcp__ruvector__hooks_route_enhanced { "task": "test" } (MCP)	spawnSync /bin/sh ETIMEDOUT	~immediate fail

mcp__ruvector__hooks_diff_analyze, mcp__ruvector__hooks_trajectory_step and other ruvector MCP tools all work in the same MCP server session, so the MCP transport itself is fine — the issue is specific to hooks_route_enhanced.

Likely cause

Looks like the MCP handler for hooks_route_enhanced is shelling out via child_process.spawnSync('/bin/sh', ...) with a timeout that is set too low (or set to 0 and rounded), so the child process is killed before it can boot. The other ruvector MCP hooks that work (diff_analyze, etc.) appear to call the underlying logic in-process, which would explain why only this one fails.

If the in-process path exists, switching hooks_route_enhanced to use it (the same code path the CLI command exercises) should fix it. If a shell-out is genuinely required, raising/removing the spawnSync timeout would unblock it.

Impact

In Claude Code, this hook is wired into the PreToolUse: Task hook chain (see .claude/settings.json from claude-flow's defaults) for enhanced agent routing. Because the MCP call always errors, this routing tier is silently disabled. The fallback mcp__claude-flow__hooks_route still works, so Task spawns are still routed — just without ruvector's AST/coverage/diff signals.

Workaround

Replace the MCP-tool hook with a command hook that calls the CLI directly:

{
  "type": "command",
  "command": "ruvector hooks route-enhanced \"\${tool_input.description}\" 2>/dev/null || true",
  "timeout": 2000
}

This bypasses the broken MCP handler and uses the working CLI path.