The Agent Skills Directory

[DATA_EXPOSURE]: The skill reads from local Chrome user data directories (e.g., ~/Library/Application Support/Google/Chrome on macOS) to find the DevToolsActivePort file required for Chrome DevTools Protocol (CDP) connectivity. While this directory contains sensitive session information, the access is used locally to establish the debugging connection.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests and executes steps from external scenario files (docs/sop/*.yaml). A malicious scenario could potentially lead to unauthorized actions within the user's active browser session.\n
Ingestion points: Scenario files in docs/sop/*.yaml and pending task lists in pending.json.\n
Boundary markers: Absent; steps are processed as defined in the input files.\n
Capability inventory: Browser control via CDP, file system writes for reports and evidence, and subprocess execution for utility scripts.\n
Sanitization: Filenames for evidence are sanitized using a whitelist regular expression, but the automation logic is dictated by the input data.\n- [COMMAND_EXECUTION]: The skill executes several commands including npx playwright test, python3 scripts/case_tags.py, and python3 scripts/to_scenarios.py to perform its tasks.\n- [EXTERNAL_DOWNLOADS]: The skill uses npx playwright install ffmpeg to download the ffmpeg dependency required for screencast recording during tests.

playwright-acceptance