The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions mandate the execution of a Node.js script (index.js) as a first action and describe using playwright-cli to attach to a browser via the Chrome DevTools Protocol (CDP).
[DYNAMIC_EXECUTION]: The index.js script dynamically loads modules from computed relative library paths (../lib/) and internal adapter paths. Additionally, the SKILL.md documentation provides a shell command that uses node -e to dynamically require a library from a path constructed using the user's home directory environment variable.
[INDIRECT_PROMPT_INJECTION]: The skill acts as an interface to an external LLM (Muse.ai), creating an attack surface where untrusted data from the external agent could influence the operator agent's actions.
Ingestion points: Muse.ai chat responses and research summaries are ingested into the agent context.
Boundary markers: The skill includes instructions to quote returned sources and use short, turn-based prompts to manage output.
Capability inventory: The skill is designed to perform shell command execution (Playwright, Node) and local file access (read and stat).
Sanitization: No explicit sanitization or escaping of external LLM content is implemented in the provided logic.

AgentChat-Muse