The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content including HTML, JSX, TSX, Vue, Svelte, and CSS files, as well as Office and PDF documents. This creates a surface for indirect prompt injection attacks where malicious content in these files could influence agent behavior.
Ingestion points: The skill analyzes a wide variety of web UI code and document formats provided by the user (e.g., in SKILL.md and references/decision-matrix.md).
Boundary markers: While the skill enforces JSON output from its sub-agents, it lacks explicit boundary markers or 'ignore' instructions for the source code being analyzed.
Capability inventory: The orchestration includes a web-issue-fixer helper that can modify source files, and uses a Node.js script for report generation.
Sanitization: There is no documented sanitization or escaping of the input source code before it is processed by the agent or its specialists.
[COMMAND_EXECUTION]: The skill is configured to execute a local Node.js script to generate its final reports.
Evidence: SKILL.md specifies rendering the report with node skills/a11y-core/scripts/render-report.mjs.

accessibility-lead