The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content including issue text, PR comments, and code comments.
Ingestion points: Files throughout the repository are scanned using rg in the scan phase (documented in references/scan.md).
Boundary markers: SKILL.md contains explicit instructions to ignore directives embedded in scanned content, such as requests to reveal secrets or alter scope.
Capability inventory: The skill can perform file writes via git and execute predefined verification commands (npm run build, npm run lint, etc.) as documented in references/fix.md.
Sanitization: No technical sanitization is described; the skill relies on instructions to ignore embedded requests.
[COMMAND_EXECUTION]: The skill executes specific shell commands for code verification and artifact generation.
Evidence: SKILL.md and references/fix.md authorize the use of npm run build, npm run typecheck, npm run lint, Vitest, and npm run generate:settings-schema.
Context: These commands are strictly limited to the project's own verification toolchain to ensure the integrity of applied hygiene fixes.

repo-hygiene