mycocapture-pipeline-durabilitylisted

# Capture Pipeline Durability and Silent Failure Detection The Myco capture pipeline can fail silently while the daemon appears healthy—sessions buffer durably but aren't ingested until manual restart. This skill covers detecting these failure modes, diagnosing root causes using four-layer architecture, and implementing resilient recovery patterns for managed daemon services. ## Prerequisites - Myco daemon running as a managed service (launchd/systemd) - Understanding of capture hook flow: client → daemon → ingestion → DB - Access to daemon logs and buffer directories (`.myco/buffer/`) - Access to `~/.claude/projects/<project>/<session>.jsonl` transcript files - Familiarity with MCP bridge architecture (agent→MCP→daemon via stdio) ## Procedure 1: Detecting Silent Capture Failures **Symptoms to watch for:** - Sessions show in `.myco/buffer/` but not in daemon logs - `/health` returns 200 OK but capture events aren't processed - Manual daemon restart suddenly processes queued events - Capture hooks time out while other daemon endpoints work **Detection steps:** 0. **Check `~/.myco/logs/launcher.log` first:** The global launcher appends a timestamped one-line record to this file on every hook launch failure (ENOENT, signal kills, path-resolution errors, binary exec errors). This is the fastest way to rule out hook-layer failure before inspecting daemon internals. If the file is absent, zero hook launch failures have occurred — the hook is reaching the daemon. If entrie