The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes repository-specific instruction files (AGENTS.md, CLAUDE.md) and configuration manifests, which represent an attack surface where malicious data could influence agent behavior.
Ingestion points: SKILL.md identifies AGENTS.md, CLAUDE.md, and various skill files as primary inputs for auditing.
Boundary markers: The skill does not explicitly define delimiters to isolate the content being audited from the agent's system instructions.
Capability inventory: The skill allows the agent to read repository files, execute git log commands, and modify file contents.
Sanitization: No specific sanitization or validation of the audited content is described before processing.- [EXTERNAL_DOWNLOADS]: The EVIDENCE.md file references research papers hosted on arxiv.org to support the auditing methodology.

context-fix