The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to perform technical audits by executing standard development and version control commands within the local repository environment.
Evidence: Utilizes commands such as git status, git diff, npm test, vitest, and supabase test db to gather evidence and verify code correctness.
Context: These operations are necessary for the primary purpose of auditing implementations and are restricted to the user's workspace context.
[PROMPT_INJECTION]: The skill includes an adversarial reasoning framework for its 'self-review' mode, which specifically instructs subagents to treat prior implementer assumptions as untrusted.
Evidence: Instructions like 'Treat prior assumptions as untrusted and require diff/test/UI evidence for every claim' enhance the reliability and security of the review process by enforcing verification.
[DATA_EXPOSURE]: The skill processes repository data, including schema files and database migrations, to verify security policies like Row Level Security (RLS).
Evidence: Includes a checklist for DB/migration review to ensure RLS/policy behavior matches the intended access model and that migrations do not introduce permission drift.
Context: This is a security-positive feature intended to identify potential data exposure risks in the code under review.

autistic-code-review