The Agent Skills Directory

[PERSISTENCE]: The setup_service.py script generates a macOS LaunchAgent plist file at ~/Library/LaunchAgents/com.clawdbot.dingtalk-bridge.plist. This allows the bridge service to automatically start at login and restart if it crashes.
[CREDENTIALS_UNSAFE]: The bridge.py script reads the file ~/.clawdbot/clawdbot.json to retrieve a gateway authentication token. While this token is used for local communication with a WebSocket server on 127.0.0.1, it involves the automated reading of stored credentials from the user's home directory.
[INDIRECT_PROMPT_INJECTION]: The skill takes untrusted text input from DingTalk outgoing webhooks and forwards it to an AI agent.
Ingestion points: The handle_request function in bridge.py parses incoming JSON payloads from the DingTalk API, extracting user message content from the 'text' field.
Boundary markers: No boundary markers or specific 'ignore' instructions are added to the user message before it is sent to the AI agent via the ask_gateway function.
Capability inventory: The bridge script has access to local network sockets and the file system (to read configuration). The downstream agent likely has broader capabilities which could be targeted.
Sanitization: The script performs minimal cleaning (regex to remove @mentions) but does not sanitize or escape the content against prompt injection attacks.
[COMMAND_EXECUTION]: The SKILL.md and README.md files contain instructions for the user to execute shell commands, including 'uv sync', 'uv run python bridge.py', and 'launchctl load' for service management.
[EXTERNAL_DOWNLOADS]: The skill uses 'uv sync' as part of its setup process, which downloads and installs Python dependencies from external registries.

dingtalk-connection