Convex Self Heal

Convex-self-heal is the loop you'd want after a prod error on a Convex backend and you're tired of babysitting the triage. It pulls errors from the sentinel table, classifies them (transient, config, or real code/schema defect), runs root-cause analysis, repairs on a branch, and certifies the fix with tsc, a migrate-rehearse run on a prod-snapshot preview where schema is involved, and a replay of the triggering input to confirm the failure is gone. It then opens a PR with the evidence attached and stops there. It never merges or deploys. The honest take: the certification step is what separates it from generic error-to-draft-PR bots, but it only works if sentinel is already capturing errors, and it will happily report "could not certify" rather than guess, which is the right call but means you'll still be doing some of the work yourself.

skills.sh installs: 41K
GitHub stars: 69
last updated: Oct 26, 2026

View on GitHub

Install

$ npx -y skills add convex/agent-skills --skill convex-self-heal --agent claude-code

Installs into .claude/skills of the current project.

Files

Gated production self-healing loop

Sentry/Datadog/Vercel can go error→investigate→draft-PR, but they treat the backend as opaque and stop at the human merge gate with an unverified diff. Convex can do the step they can't: because the error rows live in the user's own deployment and the fix can be rehearsed on a preview of that deployment, the platform certifies the fix against real invariants before anyone reviews it. This capability is the composition capstone — it wires sentinel (capture) → the findings bus (diagnose) → the fixers (repair) → migrate-rehearse/tsc/probe (certify) → a human PR (decide) → deploy-guard (promote). The human keeps the merge button; the machine does everything up to and including proving the fix works.