Social Hub

SKILL.md describes a local WeCom personal-agent that stores profiles in ~/.matchbot/chromadb, uses a local ChromaDB, and communicates with an internal matching group. The included code files, however, are a set of Claw Club shell scripts (check.sh, engage.sh, feed.sh, post.sh, register.sh, reply.sh) which interact with https://api.vrtlly.us and expect a CLAW_CLUB_API_KEY. There is no mention of Claw Club in the description, and key runtime artifacts referenced in SKILL.md (references/* files, WeCom endpoints, ChromaDB integration code) are not present. This is a clear purpose–capability mismatch.

SKILL.md instructs the agent to read/write local ChromaDB at ~/.matchbot/chromadb/, state.json, pending_matches.json, to call embedding APIs, and to send PROFILE_UPDATE and match messages to an internal group. None of the shell scripts implement these actions. Conversely, the scripts perform network calls to an external domain and read/write ~/.config/claw-club/credentials.json — paths and network endpoints that are not referenced in SKILL.md. The instructions are therefore inconsistent and give the agent broad, unspecified discretion (LLM prompts and embedding use are described but no API keys or concrete implementation are declared).

There is no install spec (instruction-only), which is lower-risk in general. However, the package includes runnable shell scripts that assume the presence of system utilities (curl, jq) and an API key. No install-time validation or dependency declaration exists, so a user could run these scripts unknowingly. Files are local (no downloads), but executing them will make network requests to an external service.

The skill declares no required environment variables or primary credential, yet the included scripts clearly expect and try to load a CLAW_CLUB_API_KEY (from env or ~/.config/claw-club/credentials.json). SKILL.md expects credentials and endpoints for WeCom and an embedding API (implicitly), but none are declared. This mismatch means the skill will either fail or prompt for/require secrets not documented — a red flag for potential credential leakage or surprise external communications.

The skill is not flagged always:true and has no install script altering other skills or system-wide settings. That said, the included scripts are designed for periodic/heartbeat use (engage.sh mentions cron/heartbeat) and will perform autonomous network calls if run. Autonomous invocation + unexpected external endpoints increases blast radius, but there is no evidence of privileged system changes.