The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data (e.g., code diffs, project plans) to generate HTML views. This creates a surface for indirect prompt injection where malicious data could influence the generated output.
Ingestion points: User-supplied plans, evidence, and code changes processed by SKILL.md and references/code-change-review.md.
Boundary markers: Instructions explicitly mandate treating supplied content as data and never as executable markup in SKILL.md.
Capability inventory: The skill generates HTML files with embedded CSS and JavaScript capabilities.
Sanitization: Instructions emphasize escaping all repository-supplied data and rendering code as text in references/code-change-review.md.
[EXTERNAL_DOWNLOADS]: The documentation mentions well-known libraries such as D3.js, htmx, and Datastar. It includes a specific "External dependency policy" in references/dependency-policy.md that prioritizes staticization and bundled runtimes over remote execution to minimize security risks. These are recognized as legitimate development tools.
[OBFUSCATION]: A base64-encoded string is used for a favicon in assets/base.html. This was analyzed as a standard PNG image for the "Quanti Pixels" brand and contains no hidden or malicious instructions.

html-artifact