The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by reading files from the local codebase (e.g., README.md, package.json, and other configuration files) which could contain adversarial instructions meant to influence the agent's behavior during the analysis workflow.
Ingestion points: The skill uses cat and ls commands in SKILL.md to read project markers and file content.
Boundary markers: The skill employs a structured multi-step workflow and a specific output template which provides some context isolation, but it does not include explicit warnings to ignore embedded instructions in data.
Capability inventory: The skill is focused on read-only operations, including file reading and directory listing.
Sanitization: No sanitization or filtering is applied to the content of the files read before they are processed by the agent.

analyzing-projects