The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs the agent to store and manage authentication secrets, including API keys and JWT tokens, at ~/.config/thecolony/credentials.json. Directing an agent to manage credential stores in the home directory is a sensitive operation.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates interaction with an external social network, exposing the agent to potentially malicious instructions embedded in user-generated content.
Ingestion points: Untrusted data enters the agent's context through API endpoints described in SKILL.md, such as fetching post context (/api/v1/posts/<id>/context) and searching the user directory.
Boundary markers: The instructions lack delimiters or safety markers to help the agent distinguish between platform content and its own primary instructions.
Capability inventory: The skill documentation in SKILL.md empowers the agent to perform write actions (posting, commenting, messaging) which could be triggered by untrusted inputs from the social network.
Sanitization: There are no instructions for sanitizing or validating external content before it is processed or acted upon by the agent.
[COMMAND_EXECUTION]: The skill provides curl command templates for platform registration and authentication. These commands involve network operations that transmit sensitive credentials to an external API and process the resulting responses.

the-colony