The Agent Skills Directory

[SAFE]: The skill is composed of purely instructional markdown files (SKILL.md and reference guides). It does not contain any executable code, scripts, or system command patterns. All external links provided are plain-text references to reputable academic papers and established technical documentation repositories.
[INDIRECT_PROMPT_INJECTION]: The skill methodology involves analyzing user-provided systems or documents ('candidates'). Ingestion points: The skill ingests untrusted data describing a domain or product for review, as referenced in SKILL.md. Boundary markers: The 'Review contract template' and specific instructions to 'separate fact, inference, and unknown' provide a conceptual framework for processing data. Capability inventory: The skill identifies no executable tools, network capabilities, or file-writing permissions. Sanitization: No specific technical sanitization is defined, but the lack of any actionable tools mitigates the potential impact of instructions embedded in the analyzed data.

review-domain