The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to resolve a target repository and read instructions from files like AGENTS.md or CLAUDE.md in SKILL.md. This allows external content to dictate agent behavior. * Ingestion points: Target repository root, AGENTS.md, CLAUDE.md, and operating documents referenced therein. * Boundary markers: The skill does not define delimiters or instructions to ignore malicious prompts within ingested files. * Capability inventory: The agent is authorized to read files and run shell commands to execute repository hooks or gates. * Sanitization: No validation of the instruction content is performed.
[COMMAND_EXECUTION]: The procedure requires the agent to manually run the target repository's hooks or gates. This pattern allows for the execution of arbitrary scripts provided by the target repository, which could lead to code execution if the repository is malicious.

repo-cosplay