The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use 'npx cypress tap' for session control, spec execution, and state inspection. These commands are local to the development environment and are the intended interface for the Cypress vendor tool.
[DATA_EXFILTRATION]: The skill accesses technical data from the application under test, including DOM source code, ARIA semantic trees, and console logs. It also describes saving command data to local JSON files ('.tap-command.json'). While these involve reading sensitive application state, they are constrained to the local environment for testing purposes.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from the application under test and processes it to make testing decisions. 1. Ingestion points: App frame content (HTML, ARIA tree) and console logs accessed via 'cypress tap' (detailed in references/reading-the-app.md and references/reading-results.md). 2. Boundary markers: None; the agent is instructed to read and interpret the app state directly. 3. Capability inventory: Local command execution ('cypress tap') and file writing (JSON logs, temporary spec files as seen in SKILL.md and references/recipes.md). 4. Sanitization: None; data is read as-is from the browser renderer.
[REMOTE_CODE_EXECUTION]: The skill facilitates dynamic code execution by instructing the agent to author temporary test scripts (e.g., 'cypress/e2e/_probe.cy.js') and execute them via the test runner. This is a standard part of the Cypress workflow for spec development and is considered a safe practice in this specific context.

cypress-tap