The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates and executes local shell scripts (e.g., dispatch.sh) to interface with the opencode CLI in 'cli' mode. It also spawns opencode processes in ACP and HTTP modes.
[COMMAND_EXECUTION]: Uses subprocess.run and asyncio.create_subprocess_exec to invoke external tools like git and opencode to perform repository operations and manage subagent sessions.
[DATA_EXFILTRATION]: Communicates with a local opencode server over HTTP/JSON-RPC on 127.0.0.1 to monitor session events and verify tool-call arguments. This communication is restricted to local network traffic and is required for the tool's core functionality.
[SAFE]: The skill implements extensive security mitigations, including fail-closed working directory verification (verify-cwd.sh), tool permission allowlists (gating bash and edit tools), environment variable key validation, and symlink-safe file rewriting in the log processing script (validate-run.sh).

dispatch-opencode