The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes specific local auditing tools in a controlled manner.
Evidence: It executes forge accessibility audit --json or fullstack-forge accessibility audit --json to gather evidence.
Safety Controls: The instructions mandate that project-native read-only checks must be inspected before execution and explicitly forbid running fetched instructions or mutating scripts as shortcuts.
[PROMPT_INJECTION]: The skill processes project source code and rendered routes, which creates a surface for indirect prompt injection.
Ingestion points: The skill reads component source code and rendered routes from the repository files.
Boundary markers: The skill instructs the agent to confirm scope and applicability decisions based on evidence before running checks.
Capability inventory: The skill can execute specific audit tools and apply safe code fixes (e.g., adding labels or semantic states) to the project.
Sanitization: No explicit sanitization of the input code is described, which is common for auditing tools.

forge-accessibility