The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires access to various shell utilities including git, pnpm, node, rg, grep, and find. These are used to perform repository audits, manage git worktrees, and apply automated code fixes.
[REMOTE_CODE_EXECUTION]: The installation mode registers a persistent execution mechanism by adding a PreToolUse command hook to the agent's configuration (.claude/settings.json). This hook executes a local TypeScript script (.claude/hooks/path-guard/index.mts) on every file edit or write operation to enforce path rules in real-time.
[DATA_EXFILTRATION]: The skill is designed to perform git push operations to the repository's origin. This capability is used to commit path fixes directly to the base branch or to create Pull Requests for review.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests and processes untrusted repository data to generate and apply code fixes.
Ingestion points: The scripts/check-paths.mts script (generated from a template) performs a recursive scan of the repository, reading .mts, .cts, .yml, Makefile, Dockerfile, and .sh files.
Boundary markers: None. The content is read and processed as raw text for regex-based matching.
Capability inventory: The skill has access to high-privilege operations including Bash, Write, Edit, and git command execution.
Sanitization: None. Data from the repository is parsed and surfaced as findings which the agent then uses to interpolate and apply code fixes based on defined patterns.

guarding-paths