The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends the installation of standard development dependencies including @playwright/test, pixelmatch, pngjs, and fs-extra from the NPM registry. It also utilizes npx playwright install --with-deps to fetch official browser binaries, which is standard behavior for this testing framework.
[COMMAND_EXECUTION]: The instructions include execution of standard CI/CD and testing commands such as npm ci, npx playwright test, and npx playwright install. These operations are local to the development environment and typical for the skill's stated purpose.
[METADATA_POISONING]: The documentation references future-dated notes (August 2026) and non-existent versions of GitHub Actions (e.g., actions/checkout@v6). These inaccuracies in the reference material do not impact the functional safety of the skill but indicate a discrepancy in metadata.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a vulnerability surface by ingesting and rendering external web content to generate screenshots.
Ingestion points: Untrusted data enters the context via page.goto() calls to external or user-defined URLs in references/ci-and-snapshot-patterns.md.
Boundary markers: The skill does not implement specific delimiters or warnings to prevent the agent from interpreting text found within the screenshotted web pages as instructions.
Capability inventory: The environment allows for file system writes to the screenshots/ directory, network access via the Playwright browser, and shell command execution via the npx utility.
Sanitization: No automated sanitization or filtering is applied to the content before it is captured and processed.

playwright-visual-testing