improve

Invoke the skill on the repository; it runs recon, fans out read-only audit subagents across nine categories, vets each finding against the actual code, then writes self-contained plan files under plans/ that an executor agent can follow without any prior context.

Each plan inlines all context (file paths, code excerpts, repo conventions), gives every step a verification command with expected output, defines hard scope boundaries and STOP conditions, and ends with machine-checkable done criteria so the executor never has to judge success subjectively.

No. It is strictly read-only on source code and only writes files under plans/. The execute variant dispatches a separate executor subagent in an isolated git worktree, then reviews its diff; it never edits code, merges, or commits to your branch.

Yes. Pass a focus argument such as security, perf, or tests to audit only that category, or use quick/deep keywords to control effort. The branch variant scopes the audit to files changed on the current working branch.

Every plan stamps the commit SHA it was written against and includes a drift check command. The reconcile variant re-verifies done plans, refreshes drifted TODOs, investigates blocked ones, and retires findings that were fixed independently.

Yes, but only with the explicit --issues flag. The skill runs gh issue create per plan using the plan file as the issue body, records each issue URL in the plan and index, and skips labels that cannot be created.