The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and process external data, including GitHub issues, pull requests, and source code. This functionality is essential for its purpose of explaining problems and features. The instructions provide a framework for mitigating risks by requiring the agent to distinguish between symptoms and causes, verify evidence against source revisions, and label proposed fixes clearly.\n
Ingestion points: SKILL.md directs the agent to read issues, requests, source/API documentation, and validation evidence.\n
Boundary markers: The narrative structure uses explicit headings (e.g., "The problem — before this PR", "The solution — after this PR") to separate different states of the code and environment.\n
Capability inventory: The skill utilizes file system access to write PDF and SVG artifacts and invokes a tool at ../../../.subharness/tools/pr-writing.ts to manage pull request descriptions.\n
Sanitization: The instructions include a specific safety policy for SVG generation, prohibiting scripts, foreignObject tags, and external resource embeds, which helps prevent the creation of potentially unsafe artifacts.\n- Tool and Library Integration: The skill suggests the use of standard, well-known libraries such as reportlab and pygments for generating formatted documents. It also integrates with a local project script (pr-writing.ts) for pull request workflows, which is a common pattern in managed development environments.\n- Verification Procedures: A comprehensive verification step is included, requiring visual inspection of rendered artifacts and checking for text extraction quality. This manual oversight step is a strong security and quality control measure.

visual-explainer