The Agent Skills Directory

[COMMAND_EXECUTION]: The video conversion helper scripts provided in references/playwright-config.md (convertToMp4 and convertToGif) use Node.js execSync to run ffmpeg commands. The file paths passed to these commands are derived from testInfo.title and only sanitized for whitespace (replace(/\s+/g, '_')). Shell metacharacters such as ;, &, and $() are not sanitized, creating a command injection vulnerability if an attacker can influence the test or feature title.
[EXTERNAL_DOWNLOADS]: The skill's CI/CD documentation in references/playwright-config.md includes commands to download and install Playwright browsers and system dependencies (npx playwright install, apt-get install -y ffmpeg). While these target standard package registries and official repositories, they represent an external dependency vector.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content in the form of existing Playwright tests and user-provided feature descriptions to generate scenarios.
Ingestion points: SKILL.md and references/scenario-guidelines.md specify reading existing tests and user stories at the planning phase.
Boundary markers: The workflow utilizes a structured Script → Stage → Shoot → Deliver process which provides natural check-points, though it lacks explicit "ignore instructions" delimiters for ingested code.
Capability inventory: The skill has access to the filesystem (via fs and execSync) and network (via Playwright's API request context and browser automation).
Sanitization: No robust sanitization or escaping is performed on the ingested content before it is interpolated into test code or shell command arguments.

director