claude-code-channels

Pass

Audited by Gen Agent Trust Hub on Jul 1, 2026

Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill provides instructions for extracting bot tokens from local configuration files (~/.claude/channels/<platform>/.env) and transmitting them to official service endpoints (api.telegram.org, discord.com) via curl. These operations are described for troubleshooting and verifying bot credentials with the respective well-known services.
  • [COMMAND_EXECUTION]: Instructions involve standard shell commands for file management (printf, mkdir, grep), process control (tmux), and dependency resolution (bun install). It also suggests using the --dangerously-skip-permissions flag when running in background sessions to facilitate automation.
  • [EXTERNAL_DOWNLOADS]: Fetches messaging plugins from the official marketplace and utilizes the Bun runtime to install required dependencies. References point to legitimate platforms and official documentation.
  • [PROMPT_INJECTION]: The skill enables an environment where untrusted data from external messaging platforms (Telegram/Discord) is ingested into the agent's context. This creates a surface for indirect prompt injection where external users could attempt to influence the agent's behavior through chat messages.
  • Ingestion points: External messages from Telegram and Discord bots (SKILL.md).
  • Boundary markers: None explicitly defined in the setup instructions.
  • Capability inventory: The agent is granted messaging capabilities and potentially broad filesystem/execution access via Claude Code.
  • Sanitization: No sanitization or filtering logic is specified in the provided setup files.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 1, 2026, 03:32 PM
Security Audit — agent-trust-hub — claude-code-channels