The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill references and attempts to execute a local JavaScript file (playwright-server.js) that is not included in the skill package. It also provides an eval command that allows the execution of arbitrary JavaScript within the browser context.
[DATA_EXFILTRATION]: The skill documentation explicitly demonstrates using the eval command to extract sensitive information, such as authentication tokens, from the browser's storage. Evidence: references/commands.md contains an example using localStorage.getItem("token").
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from web pages via commands like snapshot and text without using boundary markers or sanitization. This allows external content to potentially influence the agent's behavior. Ingestion points: snapshot, text, innerText, and innerHTML commands in references/commands.md. Boundary markers: None provided. Capability inventory: Uses the Bash tool for server communication and can execute further commands based on instructions influenced by page data. Sanitization: No sanitization or validation of page content is specified.
[EXTERNAL_DOWNLOADS]: The setup instructions involve downloading the Playwright library and browser binaries from public registries. Evidence: npm install playwright and npx playwright install chromium in SKILL.md.

playwright