The Agent Skills Directory

[SAFE]: The skill operates in a report-only mode and explicitly instructs the agent not to perform file operations, commits, or pushes when performing evaluations. This directive is clearly stated in the 'Embedded Evaluation' section of SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external online sources and user-provided artifacts ('Targets'). This creates a surface for indirect prompt injection if the retrieved content contains malicious instructions. (1) Ingestion points: Step 1 (Target resolution) and Step 2 (External research) in SKILL.md. (2) Boundary markers: Not explicitly defined for untrusted content, though output is structured into specific markdown fields. (3) Capability inventory: The skill is restricted to information retrieval and reporting; it specifically forbids file edits, commits, or pushes. (4) Sanitization: No specific sanitization or filtering of external content is mentioned.
[NO_CODE]: The skill does not include any executable scripts, binary files, or package dependencies, relying entirely on natural language instructions and the agent's existing toolset.
[SAFE]: All external research leads provided in the routing table of REFERENCE.md point to well-known technology organizations and official documentation (e.g., Anthropic, Google, Stripe, Vercel), which are established and trusted sources of information.

ask-exemplar