The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides explicit shell command templates using curl to interact with the opencode-bridge service. These commands are used to upload user files and submit task instructions, involving direct interaction with the system shell to manage network data transfer.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it is designed to ingest and analyze untrusted external data from multiple formats including Excel, CSV, PDF, and Word. Maliciously crafted instructions within these files could potentially influence the agent's behavior when it interprets the results returned by the bridge or when it performs subsequent actions like writing to Feishu documents.
[DATA_EXFILTRATION]: User-provided files and instructions are transmitted via HTTP POST requests to http://opencode-bridge:8000. While this is presented as a functional component of the OpenClaw environment, it represents the movement of user data to a network-accessible service endpoint.
[CREDENTIALS_UNSAFE]: The skill references an environment variable $OPENCODE_BRIDGE_TOKEN for authentication. While it correctly suggests using a variable rather than hardcoding a secret, the use of a bearer token over unencrypted HTTP (port 8000) for internal communication is a security consideration.

opencode-writer