improvement-agent

To automate delivery governance improvement cycles, you can use an agent that reads delivery traces, detects failure and staleness signals, and orchestrates proposal lifecycles. It generates proposals, manages challenger pre-checks, records results, and updates state.

Analyzing delivery traces to detect failure and staleness signals is handled by querying traces through a dedicated trace interface. The agent filters trace data to identify where failure signals appear and where stale activity arises within the workflow.

Yes, you can manage proposal generation and challenger pre-check workflows programmatically. The agent supports both programmatic and CLI usage, allowing you to initiate cycles, generate proposals, and record challenger results automatically.

The best way to apply approved changes to a SKILL.md file safely is through a controlled, auditable workflow. The agent enforces a no-self-modify constraint, writes proposals to the workspace, and updates state only after challenger review.

Yes, there are limitations on self-modification when generating improvement proposals. The agent enforces a no-self-modify constraint, meaning it cannot directly alter its own codebase; it only writes proposals and updates workspace state files.