Troubleshooting

Start with the page that owns the failing layer.

Troubleshooting areas

Client and server: process startup, MCP registration, HTTP/stdio configuration, auth, and routing stats.
Freshness and storage: changed files, watch states, stale snapshots, SQLite locks, and storage repair.
MCP results and recovery: zero hits, continuations, targets, handles, and proof failures.
Semantic and precise navigation: provider health, local models, degraded semantic recall, SCIP generation, and partial precision.

Verify the current process with frigg --version and the live client MCP tool list.
Call workspace when repository scope, snapshot readiness, watch convergence, or runtime exposure is uncertain.
Follow canonical next_actions and exact error codes instead of guessing a recovery.
Use the CLI for operator maintenance such as frigg index; do not invent a public MCP reindex tool.
After recovery, rerun the original producer and verify the result with a fresh source read.