Daemon does not auto-respawn after Stop-Process — leaves project without incremental file-watcher

[ardelperal]

Symptom

After killing the codegraph-vba daemon for a project (e.g. Stop-Process -Id <pid> to release the DB lock for a reindex), no new daemon is spawned for that project. The MCP server keeps answering queries (it reads the DB directly), but the incremental file-watcher is gone — codegraph-vba sync no longer auto-detects file edits.

Reproduction

1. codegraph-vba daemons — pick a project daemon, note its PID
2. Stop-Process -Id <pid> -Force (e.g. needed to release the DB lock for a forced reindex)
3. codegraph-vba daemons — observe the project daemon is gone, no replacement
4. Edit any file in the project — codegraph-vba sync <project> must be run manually; nothing else picks up the change

Expected

A new daemon spawns for the project (e.g. on the next MCP tick, or immediately on detection that the project has a .codegraph-vba/ but no daemon).

Actual

Daemon is gone until the user manually runs codegraph-vba sync or restarts the MCP server. The error message in codegraph-vba index correctly says "Stop any running CodeGraph MCP server/daemon for this project and retry" — but does not suggest the reverse (respawn).

Workaround

codegraph-vba sync <project> spawns a fresh daemon and immediately syncs.

Context

• Encountered while verifying round-5 fix (fix(extractor-vba): post-extraction resolver must cover BOTH stub:true and stub:false (round-5 amplifier of #109) #110, PR fix(resolver-vba): decline runtime-object call stubs; preserve shadow user classes (#110) #113) on consumer gestion_riesgos
• codegraph-vba upgrade --check confirms v1.7.0
• See commit c52c067 in consumer workflow repo for the full session context

/cc @ardelperal

[ardelperal]

Resuelto por PR #118 (squash-merged en main): DaemonWatchdog en src/mcp/daemon-watchdog.ts que polls cada 30s el lockfile .codegraph-vba/daemon.pid y respawna el daemon si está muerto. El MCPServer (proxy mode) arranca el watchdog para cada root servido y lo apaga en stop(). File edits ahora se detectan solos en <30s después de un Stop-Process, sin necesidad de correr 'codegraph-vba sync' manualmente.

[ardelperal]

PR #118 CI verde pero merge falló por conflicto. Reabro y reintento.

[ardelperal]

Resuelto por PR #119 (squash-merged en main): DaemonWatchdog en src/mcp/daemon-watchdog.ts que polls cada 30s el lockfile .codegraph-vba/daemon.pid y respawna el daemon si está muerto. El MCPServer (proxy mode) arranca el watchdog para cada root servido y lo apaga en stop. File edits ahora se detectan solos en <30s después de un Stop-Process.