The Agent Skills Directory

[COMMAND_EXECUTION] (LOW): The skill contains a Python script scripts/check_versions.py and instructions in SKILL.md that execute shell commands (npm info, pip show, pip index).
Evidence: The script uses subprocess.run to call npm and pip binaries on the local system.
Context: These operations are used for legitimate version checking of installed dependencies to ensure the agent is using current documentation. The inputs are generally derived from the project's own package.json or explicit agent commands.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill instructs the agent to fetch and process external documentation from URLs.
Ingestion points: SKILL.md directs the agent to use fetch_webpage or browser tools on external URLs from references/doc-urls.md.
Boundary markers: Absent in the prompt templates, but the skill's primary purpose is to read documentation, not execute instructions found within it.
Capability inventory: The agent can write files and execute local commands (via check_versions.py), but the risk of a documentation page containing a 'jailbreak' that successfully triggers these specific local tools is low.
Sanitization: Relies on the agent's underlying safety filters for web browsing.

up-to-date