The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes external data including the git working tree, repository file paths, and file content landmarks to suggest a context pack for the AI. This process creates an attack surface where malicious instructions embedded in project files or git history could potentially influence the agent when it later reads the suggested files.
Ingestion points: Task descriptions provided by the user, git repository state, and the contents of files within the project (SKILL.md).
Boundary markers: The instructions do not specify any delimiters or safety prompts to warn the AI about ignoring instructions found within the analyzed files.
Capability inventory: The skill utilizes Bash to execute the local ranking script and Read to access file structures and content.
Sanitization: There is no evidence of sanitization or filtering to prevent the agent from processing or obeying instructions that might be present in the project's source code or documentation during the ranking process.

cco-pack