The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The run-code command allows for the execution of arbitrary asynchronous JavaScript functions using the Playwright page object, which can perform actions like waiting for elements or saving downloads. Evidence found in SKILL.md and references/async-and-advanced.md.
[COMMAND_EXECUTION]: The eval command enables the execution of JavaScript expressions within the browser's context to retrieve live page data or verify input states. Evidence found in references/debugging.md and references/form-and-data.md.
[EXTERNAL_DOWNLOADS]: The skill instructs the installation of the @anthropic-ai/playwright-cli package from the NPM registry. This originates from a trusted organization and is documented neutrally.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes content and accessibility trees from external, untrusted websites. 1. Ingestion points: The snapshot and screenshot commands in SKILL.md ingest external web content into the agent's context. 2. Boundary markers: No explicit delimiters or ignore-instructions warnings are provided. 3. Capability inventory: Arbitrary code execution via run-code in references/async-and-advanced.md, file uploads via upload in references/form-and-data.md, and file-system writes via download.saveAs in references/async-and-advanced.md. 4. Sanitization: No evidence of escaping or validation of external content.
[DATA_EXFILTRATION]: The skill provides patterns for extracting sensitive browser data, such as session cookies, and saving remote files to the local machine. Evidence: await page.context().cookies() recipe for cookie extraction and await download.saveAs() for file storage in references/async-and-advanced.md.

run-playwright