analyze-trajectory

To diagnose recurring CI failure trajectories, you can dispatch sub-agents to digest large CI logs and return a concise root-cause diagnosis. This approach preserves context fidelity while preventing main context bloat.

Finding the root cause of persistent CI errors involves analyzing trajectory failures with sub-agents that produce a 1-3 sentence diagnosis. This diagnosis includes supporting evidence like run IDs and commit SHAs.

Yes, you can chunk large CI artifacts during log analysis to prevent context window issues. The process uses shared_state storage for artifacts and dispatches sub-agents to handle the chunked data.

To ensure CI log diagnosis outputs are validated and reference concrete runs, the analysis enforces strict JSON outputs and validation rules. This guarantees evidence references point to specific run IDs or commit SHAs.

You should use sub-agents to analyze CI logs instead of reading them directly when dealing with STUCK tasks or recurring failures. This offloads large log reading to prevent bloating the main context while preserving fidelity.