The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of administrative commands using the bun run admin tool. These commands perform powerful operations including banning users, deleting records, revoking versions, and modifying organizational roles within the ClawHub ecosystem.
[PROMPT_INJECTION]: The skill functions as a wrapper for user-supplied data such as slugs, handles, reasons, and email content which are interpolated into shell commands. This design presents a surface for indirect prompt injection (Category 8), where malicious data provided by a user could attempt to influence the agent's behavior or bypass the established safety rules.
Ingestion points: User-supplied slugs, handles, IDs, reasons, email subjects, and email bodies (SKILL.md).
Boundary markers: The skill instructions require the agent to show the exact command and ask for confirmation before execution, and use explicit flags for email sign-off.
Capability inventory: Subprocess execution via bun run admin, file reading for email bodies, and network operations via the underlying CLI tool's API communication.
Sanitization: No explicit sanitization or escaping logic is defined in the instructions for the interpolated shell arguments.
[DATA_EXFILTRATION]: The skill includes functionality to send emails (bun run admin -- email send). While the instructions mandate explicit user sign-off for recipient, subject, and body, this capability represents a potential vector for exfiltrating information or sending unauthorized communications if the agent's instructions are overridden.

clawhub-moderation