The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from DOM snapshots and browser console telemetry which could contain malicious instructions. Ingestion points: Artifacts located in test-results/ and trace.zip files. Boundary markers: The skill includes an explicit security guardrail (ASI06) instructing the agent to treat captured telemetry as untrusted and ignore any instructions found within it. Capability inventory: The skill utilizes execute_command, read_file, and write_file tools. Sanitization: It mandates the removal of authorization headers and session cookies from HAR files before storage.
[DYNAMIC_EXECUTION]: The skill injects a CSS string via page.addStyleTag to disable animations and transitions during test execution. This is a low-risk, deterministic use of dynamic injection for UI stabilization.
[COMMAND_EXECUTION]: The skill uses execute_command to run standard Playwright CLI tools (npx playwright test). This is consistent with the skill's primary purpose and uses well-known industry tooling.

playwright-evidence-qa