poly-gents Secret scanning response
When a secret leaks, the wrong order is: try to delete the commit first. The right order is: rotate, then purge, then trace.
When to run
- GitHub secret scanning alert.
- A scanner (gitleaks, trufflehog) finds a hit.
- A user reports a token in chat, logs, or screenshots.
- You catch it yourself before commit.
The fixed sequence
- Rotate. Immediately invalidate the leaked credential at the issuer (Stripe key in Stripe dashboard, GitHub PAT under settings, AWS key under IAM, etc.). If you can't, escalate to a human with the issuer name.
- Replace. Generate a new credential and update the live consumer (env var, secret manager, deploy pipeline). Confirm the consumer is healthy with
production-validationgate 3. - Purge from source. Force-removal (
git filter-repoor BFG) plus a forced rewrite of the affected branches. Coordinate with the team on coordinated re-clone. - Purge from artefacts. Build logs, screenshots in shared docs, screenshots in Slack threads, container layers, package registries.
- Trace. Pull access logs (last 30 days or whatever the issuer retains). Look for use that did not originate from your own infra. Document.
- Write up. File a planning issue with the timeline. If access logs show unauthorized use, treat it as an incident and run
incident-postmortem-loop.
Never
- Rotate AFTER purge. The attacker still has the live key in the window.
- Delete the leaking commit before rotating. It's already on every clone.
- Skip the trace step. Even with no evidence of misuse, the log review is part of due diligence.
Tools
- Issuer dashboards (out of band) for rotate.
agent_filesystem_search+ git history tools for purge.pagerdutyto page security on-call when the secret was production / customer-data scope.