The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data sources including source code, logs, and pull request diffs to generate technical artifacts, creating a surface for indirect prompt injection attacks.
Ingestion points: As per SKILL.md, the agent inspects code, configuration, tests, logs, and diffs provided in the development context.
Boundary markers: The instructions require the agent to separate facts from inference and explicitly label simulated or invented data, aiding in context segregation.
Capability inventory: The skill generates interactive HTML, diagrams (Mermaid, SVG), and videos using frameworks like Manim and Remotion.
Sanitization: references/html.md mandates that user-supplied text be rendered as plain text to prevent the injection of malicious markup or XSS into generated HTML explainers.
[DYNAMIC_EXECUTION]: The skill involves the runtime generation and execution of code-based artifacts to visualize system mechanisms.
references/html.md provides guidelines for creating self-contained HTML files with embedded JavaScript and CSS for user interaction.
references/video.md recommends using animation engines like Manim (Python) and Remotion (React/Node) to programmatically generate video explainers based on analyzed system data.

explain-dev