The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes and audits content from files within the .agents/skills/ directory, including skills that are explicitly identified as having third-party origins in skill-sources.json. If these audited files contain malicious instructions, they could potentially influence the agent's behavior during the audit or fixing phases.
Ingestion points: The skill reads every subfolder and file within .agents/skills/ to perform its audit functions (enumerating files, checking line counts, and scanning for specific strings like package manager commands).
Boundary markers: The instructions do not provide explicit boundary markers or directives for the agent to ignore instructions embedded within the data it is auditing.
Capability inventory: The skill possesses the ability to read, move, modify (updating pointers, replacing commands, and refreshing SHA256 digests), and delete files (specifically per-skill README.md files).
Sanitization: The skill lacks mentioned sanitization or escaping mechanisms for the content it ingests from the audited files.

update-skills-directory