Failures & incidents
Your org's searchable error library — failure clustering, documented resolutions, community validation, and AI-assisted fixes.
/failures is your organization's error library — every failure that has ever hit your pipelines, clustered by error signature, searchable, and annotated with the resolutions your team has already proven. When a workflow breaks, the answer may already be written down.
Search before you debug
The incident list searches across error logs (with a regex option) and filters by repository, workflow, and status. A developer whose workflow just failed can search the error message, land on the existing incident, and read the documented fix — instead of debugging from scratch or asking around.
Failure clustering
Runs that fail with the same error signature are grouped into one incident — a flaky Docker registry that breaks 40 runs produces one cluster, not forty tickets. Each incident carries a status: Unresolved → In Progress → Resolved (or Ignored for noise).
Documented resolutions
Every incident has a Knowledge base view where the team writes down what actually fixed it:
- Write a resolution — a titled Markdown playbook: root cause, fix steps, links (e.g., to the PR that solved it).
- Community validation — resolutions can be voted up or down, and the incident itself can be marked Validated (or returned to Needs validation), so a developer knows which fixes are proven and which are a suggestion.
- Edit and prune — resolutions stay editable, so the playbook sharpens over time.
Solve with AI
No documented fix yet? The Solve with AI drawer generates a root-cause hypothesis and a draft resolution in one click — highlighting the log lines and diffs that most likely introduced the failure. Apply the draft straight into the resolution form, edit it, and save it for the next engineer. It is the same engine behind the AI agent.
The three incident views
| View | Purpose |
|---|---|
| Investigation | Timeline, step durations, runner specs, and raw error streams for the failing run. |
| Knowledge base | Documented resolutions — write, vote on, and validate the fixes that worked. |
| War room | Focused incident-response view for outages that block production. |
Failure analytics
MTTR (mean time to resolution), recurring-error trends, and a flakiness index per workflow — the inputs for regressions and auto-retry rules.
Documented resolutions feed the AI agent's similar resolutions tool — every fix your team writes down makes the next engineer's incident (and the next AI hypothesis) faster.