Codex Monitor
Use this skill when:
- setting up recurring ops summaries
- triaging what changed across builds, deploys, PRs, and incidents
- creating or updating Linear issues from operational signals
- deciding whether a separate heal agent should be triggered
Goal
Convert noise into a small operating queue:
- what broke
- what changed
- what was fixed
- what needs ownership now
Inputs
Gather from as many of these as exist:
- latest Sentry incidents
- Vercel deployments, build logs, runtime logs
- GitHub PRs, CI failures, review friction
- Linear backlog/in-progress blockers
- local reports under
dev/reports/
Workflow
- Read the current project context and recent reports.
- Collect live signals from connected systems.
- Group duplicates into one operating item.
- Score severity by business impact, not just technical noise.
- Create or update Linear issues for actionable work.
- Produce a concise summary:
- urgent now
- important but not urgent
- shipped/fixed since last check
- blocked/waiting
- Decide whether heal should be invoked.
Trigger rules for heal
Start a separate fix flow only if:
- issue is critical/high
- fix scope is bounded
- verification path exists
Otherwise:
- write/update Linear
- keep it in the summary
Linear policy
Use Linear for:
- production incidents that require ownership
- deploy failures with follow-up work
- repeated flaky behavior
- security/performance findings that are not instantly resolved
Do not create Linear issues for:
- pure duplicates
- resolved transient noise without follow-up value
Summary format
Keep it short and ranked:
- Critical
- High
- Medium
- Fixed/cleared
Each item should include:
- source
- impact
- owner if known
- Linear link if created
Separation of concerns
This skill reports and routes. It does not implement fixes.