MCP daemon returns internal error with no message (multiple independent reports during review round)

[ohdearquant]

During the 2026-06-10 paired review round, at least three independent codex review sessions hit khive MCP write/read failures of the form:

• "khive MCP write-back failed: daemon returned an error without a message"
• "mcp__khive__request for recall/suggest/stats: failed with daemon error and no message" (PR feat(runtime): write-time secret detection gate — hard block with masked reason (#76) #83 round-2 session, ~2026-06-10T21:02 EDT)

Symptoms:

• Affects multiple verbs (recall, suggest, stats — i.e. not verb-specific)
• Error carries NO message, so the failure dimension is invisible to the caller
• Reported from codex sandbox sessions running in khive worktrees (daemon shared across sessions)

Two distinct problems to investigate:

1. Whatever the underlying daemon error is (possibly contention from many concurrent sessions, or stale daemon state during the review round's builds + make local cycles)
2. The error path that swallows the message — an internal error must carry its cause to the client. Fail-closed AND say why (same standard as the reindex fail-closed work in fix(db): backfill FTS on reindex — close coverage gap for pre-existing notes #88).

Label: bug. Found during the codex+gemini paired review round; review sessions used local code paths as fallback so verdicts were unaffected.