whiteguo233 / verify

Verifying OpenBiliClaw backend changes end-to-end

Port 8420 is the user's production backend — never restart or reuse it. Spin up
throwaway serve-api instances on spare ports (846x) against isolated project
roots instead, and always kill them when done.

Recipe

1. Isolated project root: mktemp -d /tmp/obc-verify.XXXXXX, put a minimal
   config.toml + empty data/ inside, point OPENBILICLAW_PROJECT_ROOT at it.
   data_dir = "data" resolves against that root.

2. Craft the backend state you need via config/data, not mocks:

   • Configured but uninitialized: any provider section with a dummy
     api_key (registry build does not call the network at startup).

   • Degraded (llm_registry_unavailable): all provider keys empty and
     launch with `env -u OPENAIAPIKEY -u ANTHROPICAPIKEY -u GOOGLEAPIKEY
     -u GEMINIAPIKEY` — ambient keys silently un-degrade it.

   • Initialized (profile ready): copy the real data/memory/soul.json into
     the tmp root's data/memory/ before launch. is_profile_ready() ==
     "soul memory layer has data", loaded at startup (don't hot-drop the file
     into a running instance and expect it to flip).

3. Launch (background, stdin must be closed or it hangs):
   `bash
   OPENBILICLAWPROJECTROOT=/tmp/obc-verify.XXX/a \
   nohup .venv/bin/openbiliclaw serve-api --host 127.0.0.1 --port 8461 \

   /tmp/obc-verify.XXX/a/server.log 2>&1 < /dev/null &

    Ready in ~5-8s. Use the main repo's `.venv/bin/openbiliclaw`.
   

4. Probe with curl --noproxy '*' — the local proxy/TUN hijacks loopback
   otherwise. curl -D - for redirect headers, /api/health to confirm the
   state you crafted (status, profile_ready), server.log for the
   "started in degraded mode" line.

5. Clean up: kill the PIDs, confirm nothing listens on the ports
   (lsof -nP -iTCP:846x -sTCP:LISTEN), rm -rf the tmp roots. Zombie temp
   serve-api instances have corrupted real user data before — never skip this.