The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs behavioral instructions such as 'AUTONOMOUS MODE' and 'Do NOT ask questions' to suppress standard agent interaction and feedback loops.
[COMMAND_EXECUTION]: The skill performs several automated shell operations, including installing Node.js development dependencies (npm install -D), executing testing frameworks (npx playwright, npx jest, flutter test), and running Lighthouse audits (lhci autorun).
[REMOTE_CODE_EXECUTION]: The skill dynamically generates functional test code (e.g., Playwright and Jest test files) based on discovered project routes and components, then proceeds to execute this code in the local environment. This runtime code generation follows a templated approach for testing purposes.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the processing of untrusted data provided via the $ARGUMENTS variable.
Ingestion points: $ARGUMENTS parameter in SKILL.md used to define test scope.
Boundary markers: No delimiters or safety warnings are present to isolate user-supplied arguments from the internal instruction logic.
Capability inventory: The skill has broad capabilities including file system writes (test generation), package installation, and execution of arbitrary testing commands via npx and lhci.
Sanitization: There is no evidence of sanitization or validation of the input arguments before they are used to influence the testing logic or code generation.
[COMMAND_EXECUTION]: Implements a telemetry logging mechanism that writes execution metadata to a persistent file on the host machine (~/.claude/projects/skill-telemetry.md).

accessibility-test