The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to read and analyze arbitrary source code and configuration files from a repository to generate markdown lists (targets.md and checklist.md). Malicious instructions or adversarial patterns embedded within the analyzed files (such as in code comments, documentation, or string literals) could attempt to manipulate the agent's summarization process or influence the behavior of the codex-consultation tool used for final review.
Ingestion points: The skill reads a wide variety of untrusted files from the repository, including package.json, Gemfile, go.mod, controller files, and middleware definitions (SKILL.md).
Boundary markers: The instructions do not specify the use of clear delimiters or instructions to ignore embedded prompts when interpolating content from the analyzed repository into the generated markdown files.
Capability inventory: Across its operations, the skill is capable of reading any file within the repository, writing local files to the .dev/codepatrol/ directory, and invoking an external consultation tool (codex-consultation) with the aggregated content (SKILL.md).
Sanitization: There is no description of escaping, filtering, or validating the content read from the repository before it is formatted into markdown or passed to the subsequent LLM-based review tool.

codepatrol-setup