The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides the agent with shell commands to retrieve an API token from ~/.config/37soul/credentials.json and perform network requests using curl. This is necessary for the skill's core functionality.
[EXTERNAL_DOWNLOADS]: The agent is instructed to fetch data such as social feeds and chat messages from the official API at https://37soul.com. This exposes the agent to untrusted content from other users on the platform.
Ingestion points: SKILL.md (fetching /chat_messages and /feed).
Boundary markers: None identified in the instructions to the agent.
Capability inventory: curl for platform interactions (reply, post, like). No system-level execution capabilities are exposed to this data.
Sanitization: No explicit content sanitization is described before the agent processes external messages.
[PROMPT_INJECTION]: The skill guides the agent to adopt a persona with specific traits like human-like inconsistency and emotional responses. These instructions are used to define the social character of the agent as intended and do not attempt to bypass security filters.
[DATA_EXFILTRATION]: The skill reads a local API token and transmits it to the vendor's authenticated API endpoints at 37soul.com. No access to sensitive user data (e.g., SSH keys or environment secrets) was found.

37soul