The Agent Skills Directory

[COMMAND_EXECUTION]: The included Python script scripts/find_skills.py performs shell command execution by calling the npx skills utility to list, add, and verify skill installations.
[REMOTE_CODE_EXECUTION]: The skill is designed to download and install executable code from the volcengine/volcengine-skills repository. This behavior is consistent with the skill's primary purpose as a marketplace installer for the vendor's ecosystem, though it facilitates the introduction of new executable code into the agent's environment.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to analyze summaries and keywords from the references/catalog.json file to select capabilities. This creates a surface where metadata from the catalog could potentially influence agent logic.
Ingestion points: The references/catalog.json file contains metadata that the agent is instructed to read and interpret (specified in SKILL.md).
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are provided for processing the catalog data.
Capability inventory: The script scripts/find_skills.py contains subprocess.run calls capable of executing code and modifying the system state.
Sanitization: No specific sanitization or validation of the natural language descriptions is performed before the agent processes them.

volcengine-find-skills