The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes arbitrary source code from repositories to generate diagrams and documentation. This ingestion of untrusted data creates an attack surface for indirect prompt injection if the code contains malicious instructions intended to manipulate the agent's reasoning during the analysis phase. \n- Ingestion points: SKILL.md Step 1 (reading code, pull requests, and diffs) and scripts/build.py (reading source files to populate code panes). \n- Boundary markers: The skill instructions mention manually escaping HTML characters for output but do not define specific delimiters or instructions for the agent to ignore embedded commands within the processed source code. \n- Capability inventory: Local file system read access (build.py), browser automation (check.mjs), and the ability to execute local build/check scripts. \n- Sanitization: Output text is HTML-escaped to prevent rendering issues, but no semantic filtering of input code is present to mitigate prompt injection. \n- [EXTERNAL_DOWNLOADS]: The skill requires the playwright Node.js package for auditing generated explainer pages. Playwright is a standard and well-known browser automation library developed by a recognized organization. \n- [COMMAND_EXECUTION]: The workflow involves executing local Python and Node.js scripts to transform code into interactive explainers and to perform visual audits through automated browser screenshots.

visual-explainer