The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external text files, creating a surface for indirect prompt injection if those files contain instructions intended to manipulate the agent's behavior.
Ingestion points: The skill reads AGENTS.md and the root README.md as part of its auditing workflow.
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded commands within the ingested text, although the line-by-line audit process naturally restricts the scope of processing.
Capability inventory: The skill is strictly limited by the disable-model-invocation: true configuration, preventing it from calling external tools. It possesses only the ability to read and write to the local AGENTS.md file; it has no network, shell, or dynamic execution access.
Sanitization: No sanitization of the input text is performed before it is processed or used to generate the rewritten document.
[SAFE]: The skill serves a purely analytical purpose for documentation maintenance. Given its lack of executable capabilities and its focus on high-level principle auditing without performing builds or tests, it does not pose a security threat to the host environment.

audit-agents