Troubleshooting

Common issues and how to resolve them. If none of these match, use the Feedback button in your dashboard.

My PR didn't get a comment

Cause: the repo is disabled on your dashboard, webhook delivery failed, or the PR only touches unsupported file types.

Fix: confirm the repo toggle is on. Then open the installation at github.com/settings/installations and check Recent Deliveries for failed webhooks. PRs with no JS, TS, TSX, or Python files still get a review, just without structural detail.

The bot comment didn't update on a new push

Cause: Senix updates the existing comment in place rather than posting a new one, so GitHub does not bubble it to the top of the thread.

Fix: refresh the PR page. If the comment body still looks stale, analysis for the latest push may still be running — it completes within ~60 seconds.

I got an unexpected risk level

Cause: risk level is calibrated by impact-if-shipped-wrong, not by diff size. A one-line change to pricing logic is high; a 500-line refactor with no behavior change is low.

Fix: check the risk flags on the comment — they explain the level. If it still looks wrong, send the PR URL via the Feedbackbutton and we'll fold it into the next eval run. See the risk flag reference.

MCP tool not appearing in my IDE

Cause: the MCP config is malformed, in the wrong file, or the IDE has not reloaded it.

Fix: verify the JSON is valid and the url is https://senix-chi.vercel.app/api/mcp. Fully restart the IDE so it re-reads the config. Check your IDE's MCP panel for a connection error.

MCP returns Unauthorized

Cause: the token is missing, mistyped, revoked, or the Authorization header is not formatted as Bearer <token>.

Fix: go to Dashboard → MCP tokens, generate a fresh token, and paste the whole sk_mcp_… string. Tokens are shown only once — if you lost it, generate a new one and revoke the old.

Analysis took longer than 60 seconds

Cause: very large diffs, or the LLM provider is slow or rate-limited.

Fix: large PRs are capped to keep cost predictable — splitting a huge PR into smaller ones both speeds up analysis and produces sharper summaries. If a normal-sized PR consistently times out, report it via Feedback.