Commit

Binding sources

• AGENTS.md — if the change touches risk/execution/AI/learning, the commit message should not imply live enablement or L7 bypass.
• docs/FINAL_SYSTEM_VISION.md — large architectural commits should name affected layers when relevant.

Purpose

Use this skill to convert the current worktree into a clean operator-facing summary before committing.

It should understand the current repo shape:

• runtime bot changes
• experiment and scoped-trial changes
• operator panel and autonomous research workflow changes
• skill-system and documentation-only changes
• local state, generated logs, build artifacts, and secrets that must not be committed

Use When

Use this skill when:

• a scoped task is complete and ready for commit
• you want one clear commit message instead of a vague summary
• the diff touches risky areas such as app/risk, app/execution, app/operator_panel.py, scripts, or .env.example

Do Not Use When

Do not use this skill when:

• validation has not run yet
• the worktree still contains unrelated exploratory edits
• the requested output is a review rather than a commit-oriented summary

Required Checks

• Group changes by operational purpose, not by file list alone.
• Call out risky edits in risk, execution, operator actions, config, startup, or secrets handling.
• Verify generated files, local data, logs, dist/, and .env are not staged.
• State what validation ran and what still needs to run.
• Propose one semantic commit message that matches the actual scope.

Expected Output

• Change Summary: grouped by feature or workflow.
• Unsafe Changes: warnings or none.
• Validation Status: exact checks run and remaining gaps.
• Suggested Commit: one commit message.