Set up TRW for opencode

Drop TRW into opencode’s command-first terminal workflow — your slash commands become TRW skills, your agents become TRW specialists, and your instructions persist across sessions.

Prerequisites

opencode installed and on your PATH.
Python 3.10+ (3.11+ recommended) and pip 22+.
A git repository you can write to.

Install

curl -fsSL https://trwframework.com/install.sh | bash

When the installer prompts for target clients, select OpenCode. The bootstrap installs trw-mcp and writes the opencode surface into the repo.

What the installer writes

opencode.json — opencode’s root config. Registers trw-mcp as an MCP server and wires the command and agent directories.
.opencode/INSTRUCTIONS.md — durable instructions opencode reads every session. Encodes the TRW tool lifecycle so it survives context compaction.
.opencode/commands/ — TRW’s slash-skill catalog translated into opencode commands (sprint init, PRD grooming, project health, etc.).
.opencode/agents/ — specialist agents from the TRW roster mapped onto opencode’s agent surface.
.trw/config.yaml — repo-local TRW config listing opencode in target_platforms.

Run your first session

From the project root, run opencode. On startup it loads opencode.json and .opencode/INSTRUCTIONS.md, which tell the model to begin with trw_session_start().

trw_session_start() — loads prior learnings and any active run state.
Use the TRW commands in .opencode/commands/ the same way you already use opencode commands. Slash-discoverable, same behavior as the Claude Code skills.
Record discoveries with trw_learn(summary, detail).
Finish with trw_build_check() then trw_deliver().

Differences from Claude Code

No hooks directory. opencode doesn’t host TRW’s hook automation. Session-start, pre-compact, and deliver ceremony are enforced through instructions and commands instead of event triggers.
Commands replace skills. The slash-skill catalog lives at .opencode/commands/. Names and behaviors mirror the Claude Code skills so muscle memory carries over.
Agent teams are command-driven. Agents live in .opencode/agents/, but orchestration is invoked through commands rather than a native team API.
One config file, not three. opencode.json carries MCP, command, and agent wiring — there is no separate .mcp.json.

Common problems

Next steps

Core concepts

Sessions, runs, learnings, and the ceremony loop that TRW enforces on every client.

See the flagship profile

Claude Code is the reference integration — use it to understand the full surface.

Workflows overview

The six-phase lifecycle your opencode sessions will run through.