The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, including PR titles, branch names, and code diffs, to generate HTML reports. While this creates a vulnerability surface, the skill includes mandatory sanitization instructions.
Ingestion points: PR metadata (title, author, branch) and file diffs provided by the user or codebase.
Boundary markers: No specific boundary markers are defined for the agent's prompt context when interpolating this data.
Capability inventory: The skill can write HTML files to disk, invoke the html-skills:html-skills-listen tool, and use the Artifact tool for publishing.
Sanitization: The instructions explicitly mandate using textContent and createElement for DOM construction and strictly forbid innerHTML to prevent XSS attacks.
[DYNAMIC_EXECUTION]: The skill generates standalone HTML files containing embedded JavaScript for interactivity, clipboard operations, and state persistence via localStorage.
Logic: The generated code includes a handshake mechanism (window.__CLAUDE_SUBMIT_URL__) provided by a local tool to facilitate communication between the artifact and the agent environment.
State Isolation: Instructions require a unique per-artifact prefix for localStorage keys to prevent cross-artifact data access.

html-code-review