Live run reliability and recovery
Codelit validates its shipped Agent Teams and read adapters before managed work begins. These checks construct the same bounded execution plans but do not contact a provider, call a model, allocate a browser, or write customer data.
Know what the readiness check covers
The production check validates every flagship Team, committed Sample fixture, connected-app read executor, and native provider-pack read adapter. It also verifies that every consequential flagship action remains behind its declared approval gate.
- Provider requests built: zero sent.
- Model calls and browser allocations: zero.
- Customer and provider writes: zero.
Recover without losing the workflow
If a readiness check fails, hosted and managed Live runs pause before customer work. The Team, run brief, connections, and history remain saved. Free Sample and eligible local or in-tab review paths remain available while Codelit restores the affected live boundary.
- Open Notifications or the run terminal for the bounded failure state.
- Keep using Sample or local review when the outcome can be proven without managed execution.
- Retry Live only after the status clears; a paused attempt does not authorize or replay an external write.
Trace IDs, not customer content
Runtime diagnostics correlate only allowlisted route, run, deployment, provider, operation, reservation, approval, receipt, status, duration, and error-code fields. Prompts, messages, emails, URLs, credentials, source excerpts, and provider response bodies are not accepted by the runtime log record.
Use the operator health receipt
The no-store `/api/canary/agent-teams` route returns the current check counts, duration, enforcement state, and zero-side-effect receipt. It contains no credentials, customer content, source data, or constructed provider request.