The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the local utility script scripts/check-skill-paths.sh to verify that all repository paths referenced in the newly generated auditing skill are valid.
[DYNAMIC_EXECUTION]: The skill generates shell commands for code searching and auditing based on the target repository's context and then executes these commands in a trial run to verify the generated skill's functionality.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from repository-local documentation and history to inform the content of the generated skill, creating a vulnerability surface where malicious instructions in the repository could influence the agent's generated guidance.
Ingestion points: Reads files including AGENTS.md, CLAUDE.md, CONTRIBUTING.md, REVIEW.md, SECURITY.md, and various documentation files within the docs/ directory of the repository being analyzed.
Boundary markers: No explicit delimiters or boundary markers are utilized during the data ingestion phase to prevent the agent from following embedded instructions in the documentation.
Capability inventory: The generated skill utilizes rg and git ls-files for searching, and executes repository-defined commands (such as npm run lint, knip, or bun run check) during the trial validation phase.
Sanitization: The skill does not implement explicit sanitization or filtering of the ingested content before interpolating it into the generated skill's instructions.

create-code-scans-skill