Guard · Incidents

Incidents

Every prod alert, blamed back to the change that caused it. Each new prod-adapter event (Sentry / Vercel / PostHog / DB slow-log) is correlated to the introducing commit and the AI session that wrote it, ranked by confidence. Correlation runs LLM-free.

Where to find it

  • Desktop: http://localhost:4000/, then Guard → Incidents in the sidebar.
  • Hosted: repoops.ai/team/incidents
  • Keyboard: ⌘ K, then type “Incidents”.
  • API: GET /api/incidents?repo=<id>&limit=<n>
  • MCP tool: get_prod_incidents

What it does for you

Every incident names its likely cause.The correlator blames the implicated file line back to its introducing commit, then matches that commit to the AI session that wrote it. Each incident card names the commit, the author, the session, and links to the source alert.
Confidence-banded, never a guess dressed as fact.Exact attribution needs a Session-Id commit trailer; without one, the correlation is banded by confidence (strong, likely, weak) instead of asserted.
Feeds the accountability ledger and blast-radius reads.Incidents are one of the signals the Accountability ledger fuses into its per-loop story, and back the agent blast-radius MCP tool.

Configure

Opt-in per source. Sentry, Vercel, PostHog, and DB-slow-query ingest are each off by default, with redaction at ingest.

Read more

Last updated