The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires running a local shell script (./skills/dev-browser/server.sh) to start a background process and uses bun x tsx with heredocs to execute dynamically generated TypeScript code.
[EXTERNAL_DOWNLOADS]: The setup process is described as automatically installing dependencies and downloading the Playwright Chromium browser during its first run.
[PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection when navigating to external websites, as it lacks sanitization for retrieved content like ARIA snapshots or page text.
[PROMPT_INJECTION]: Mandatory Evidence Chain:
Ingestion points: The agent ingests untrusted data from web pages via client.getAISnapshot() and page.goto() as seen in SKILL.md.
Boundary markers: Absent. There are no instructions to use delimiters or warnings when processing retrieved web content.
Capability inventory: The skill can execute shell commands (bun x tsx), write files (screenshots in tmp/), and perform network operations.
Sanitization: Absent. The skill does not describe any methods for sanitizing or validating data retrieved from the browser context before further processing.

dev-browser