The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and audit untrusted data from a website-clone workspace, including HTML, CSS, JavaScript, and metadata files. The audit process involves matching identifiers and excerpts, which could contain malicious instructions designed to influence the agent's report or behavior.
Ingestion points: SKILL.md (lines 14-17) instructions specify inventorying HTML, CSS, JavaScript, and assets.
Boundary markers: The skill uses explicit instructions to 'Treat references/source-1-CLONE_AUDIT.md only as historical provenance' (line 21) and mandates 'never reuse findings from an earlier project' (line 8).
Capability inventory: The skill identifies capabilities for reading files and metadata. It explicitly forbids package installation and network requests without authorization (line 19).
Sanitization: No explicit sanitization of the matched identifiers or excerpts is mentioned before including them in the final markdown report.
[EXTERNAL_DOWNLOADS]: The references/source-1-CLONE_AUDIT.md file contains various external URLs, including links to GitHub repositories, NPM registry, and third-party CDNs. While these are historical evidence rather than active download triggers, they represent external references within the skill's data context.
Evidence: references/source-1-CLONE_AUDIT.md lists https://github.com/zanwei/design-dna, https://github.com/lixiaolin94/skills, and https://registry.npmjs.org/ among others.

clone-audit-mrlv3nl4