The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute various shell commands using vp (Vite+ CLI) and pnpm. Examples include linting files (vp dlx comark-lint), rendering content (vp exec comark-annotate), and installing dependencies (pnpm add -D comark-annotate).
[DYNAMIC_EXECUTION]: The skill provides a complete Node.js script (record.mjs) in its documentation and instructs the agent to save and execute it locally. This script uses child_process.execFileSync to run ffmpeg and performs file system operations like mkdirSync and writeFileSync to generate video clips of UI interactions.
[INDIRECT_PROMPT_INJECTION]: The skill processes external web content by navigating to URLs provided by the user to capture screenshots or recordings. This creates an attack surface where a malicious webpage could attempt to influence the agent's behavior.
Ingestion points: The node record.mjs <url> command and Playwright-based screenshot steps ingest content from user-specified URLs (e.g., branch previews or deployments).
Boundary markers: None identified in the prompt templates or scripts to isolate external content from instructions.
Capability inventory: The skill has the capability to write to the local filesystem (writeFileSync), create directories (mkdirSync), and execute subprocesses (execFileSync, vp, pnpm).
Sanitization: There is no explicit sanitization of the external web content or the URL before processing, although Playwright provides standard browser-level isolation.

before-after-evidence