The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the project repository to generate or audit durable agent instructions.
Ingestion points: The skill reads project roots, worktrees, manifests, source structure, and existing instruction files (e.g., AGENTS.md, CLAUDE.md) as specified in the SKILL.md "Preparation" and "Audit and create" sections.
Boundary markers: While SKILL.md mentions distinguishing verified claims from unverified ones, it lacks explicit technical delimiters or instructions to ignore potential commands embedded within the analyzed project files.
Capability inventory: The skill utilizes codebase-xray:xray-method and instructions-auditor, and it possesses the capability to modify durable project files based on the analyzed content.
Sanitization: There is no evidence of sanitization or filtering of the ingested project data to remove or neutralize malicious prompt injection patterns before they are processed by the agent.

instructions-method