Troubleshooting

Most issues come down to one of three things: the worker isn't running, an item didn't clear the gate, or a tool isn't on the PATH. Here's how to tell which.

My feedback is stuck on “Awaiting triage”

“Awaiting triage” means no worker has claimed the item yet — almost always because the worker is offline. Open Account → Local worker. If the pill is grey, start the worker. If it's running but items still aren't moving, run node fdbck-worker.mjs doctor to confirm it can reach fdbck and sees your repos.

The worker won't start

Run doctor first — it points at the exact failing check.

node fdbck-worker.mjs doctor

The usual culprits are a missing tool on the PATH or an unauthenticated one. Confirm each: node --version (≥ 18), git --version, gh auth status, and that your agent CLI (claude, codex, or gemini) is logged in. Also check that each path in worker.json points at a real checkout.

No PR was opened

Usually the item was triaged but didn't clear the Auto-PR gate — its confidence or severity came in under your thresholds — so it's waiting for you in the queue rather than getting a fix. If Review before opening is on, the fix exists as a parked diff awaiting your approval (the item shows Awaiting review) — approve it and the PR opens. Otherwise cleared items open as draft PRs, so also check GitHub's draft list, not just open PRs. If it should have opened, check the item's run log for what the worker did. See Automation & review for how the gate works.

How do I revoke a token, or what if one leaks?

Go to Account → Local worker, revoke the worker token, and generate a new one. Then update token in ~/.fdbck/worker.json and restart the worker. A revoked token stops working immediately. The widget key (wk_…) is public and separate — leaking it only lets someone submit feedback, not run a worker. You can rotate it from Settings.

Does fdbck see my code or my Claude credentials?

No. All of the model work and all of the git work happen on your machine, with your own agent (Claude, Codex, or Gemini) and your own gh/git credentials. fdbck stores no model keys. It only holds the feedback itself and the state of the queue — not your source, your history, or your logins.

How much does it cost?

Triage and fixes run on your own agent subscription (Claude, OpenAI, or Google) through your local CLI, so there's no per-fix charge from fdbck — model usage is billed to you by your provider.