manage-channels

Warn

Audited by Gen Agent Trust Hub on Aug 30, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONPRIVILEGE_ESCALATION
Full Analysis
  • [DATA_EXPOSURE]: The skill instructs the agent to access sensitive configuration files and the application database to retrieve credentials and role information.
  • Evidence: The instructions explicitly direct the agent to "check .env for channel tokens".
  • Evidence: The skill provides SQL queries to read from data/v2.db, targeting tables such as user_roles and messaging_groups which contain sensitive platform IDs and user authorization levels.
  • [COMMAND_EXECUTION]: The skill relies on executing local scripts and CLI tools to perform its primary functions, providing the agent with shell execution capabilities.
  • Evidence: Use of pnpm exec tsx scripts/q.ts to execute arbitrary SQL queries against the local database.
  • Evidence: Use of the ncl CLI tool for creating messaging groups and wiring configurations.
  • Evidence: Use of pnpm exec tsx setup/index.ts to perform multi-step registration and pairing processes.
  • [INDIRECT_PROMPT_INJECTION]: By wiring the agent to external messaging platforms (Telegram, WhatsApp, Slack, etc.), the skill establishes an attack surface where untrusted data from external users could influence agent behavior.
  • Ingestion points: External messages arriving via configured platforms (e.g., WhatsApp, Telegram groups).
  • Boundary markers: None specified in the instructions to distinguish between operator commands and data from the messaging channels.
  • Capability inventory: The skill has the ability to execute local scripts (tsx), interface with a custom CLI (ncl), and perform write operations on the application database.
  • Sanitization: No sanitization or validation protocols are described for processing inbound message content before it reaches the agent context.
  • [PRIVILEGE_ESCALATION]: The skill describes and provides tools for managing the internal administrative roles of the application.
  • Evidence: Instructions for querying and displaying users with owner and admin roles.
  • Evidence: References to grantRole() and init-first-agent logic which handles the promotion of users to the owner role.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 30, 2026, 02:00 AM
Security Audit — agent-trust-hub — manage-channels