The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides automated commands to install a persistent background service on macOS using launchd (feishu-codex-bridge service install), allowing the software to maintain a persistent presence and run automatically on the host machine.
[COMMAND_EXECUTION]: The tool’s core functionality involves spawning local subprocesses to execute commands like codex exec and lark-cli using inputs received from remote chat sessions.
[EXTERNAL_DOWNLOADS]: The installation instructions require downloading the source code from a non-trusted third-party GitHub repository (QQQingyu/feishu-codex-bridge) and installing several global NPM packages.
[REMOTE_CODE_EXECUTION]: The skill creates a direct remote-to-local execution surface where instructions sent via an external chat platform are translated into executable code tasks on the local system.
[DATA_EXFILTRATION]: Through its integration with lark-cli, the bridge can access and potentially export sensitive data including cloud documents, calendars, and chat history, while also downloading file attachments from the chat platform to a local media directory.
[PROMPT_INJECTION]: The skill is inherently vulnerable to indirect prompt injection because it processes untrusted inputs from chat messages and attachments to drive its execution engine.
Ingestion points: The skill reads and processes all incoming messages and file attachments from Feishu/Lark chat threads.
Boundary markers: No delimiters or "ignore instructions" markers are used to separate chat data from the instructions provided to the underlying execution engine.
Capability inventory: The skill possesses extensive capabilities including local file system access, shell command execution, and network communication via APIs.
Sanitization: There is no evidence of input validation or sanitization to prevent malicious instructions embedded in chat messages from being executed by the bridge.

feishu-codex-bridge