The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the Membrane CLI via npm install -g @membranehq/cli. This is a vendor-owned package from the official NPM registry and is used for authentication and API proxying.
[COMMAND_EXECUTION]: The skill uses various membrane CLI commands (e.g., membrane login, membrane connect, membrane action run) to interact with the CompanyCam service. These commands are restricted to the vendor's platform environment.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it processes user-generated content (comments, photo descriptions) from CompanyCam.
Ingestion points: Data enters the context through membrane action run and membrane request calls that fetch projects, photos, and comments (SKILL.md).
Boundary markers: None identified; external data is not wrapped in markers to distinguish it from instructions.
Capability inventory: The skill can perform write operations like create-project, delete-photo, and archive-project via CLI commands.
Sanitization: No explicit sanitization or filtering of fetched API data is documented.
[METADATA_POISONING]: The 'CompanyCam Overview' section in SKILL.md contains hundreds of irrelevant items, including lists of Kepler planets, historical philosophers, and international organizations. This appears to be low-quality documentation or a generation hallucination but does not contain executable malicious instructions.

companycam