Skip to main content
TRW
Skip to content
TRWQuickstart — Install in Under Two Minutes

Docs

Quickstart

Review the current access flow, run the installer, choose your client surfaces, verify the setup, and start your first retained-context workflow. The goal is simple: wire the repo once, then preserve the decisions and evidence later sessions would otherwise lose.

Fast path

  1. 01

    Open the project root

    Start in the repository you want TRW to configure. Your source tree and existing client stay in place.

  2. 02

    Run the bootstrap

    One shell command installs trw-mcp and starts guided setup.

  3. 03

    Sign in and choose clients

    Approve the device when prompted, then select every AI surface this repository already uses.

  4. 04

    Verify and start

    Confirm the generated files, open one configured client, and start a small real session.

Prerequisites

Requirement
Python
Version
3.10+ (3.11+ recommended)
Install
brew install python / apt install python3
Verify
python3 --version
Requirement
pip
Version
22.0+
Install
python3 -m pip install --upgrade pip
Verify
python3 -m pip --version
Requirement
git
Version
2.30+
Install
brew install git / apt install git
Verify
git --version
Requirement
AI coding client
Version
latest stable available to you
Install
Use one of the installer-supported clients listed below
Verify
Run your client’s version command or open the repo in the client you plan to use

Installation

Run one command from the project root. The bootstrap installs trw-mcp, opens device sign in when needed, and starts the guided setup. CI can pass an existing API key instead.

shell bootstrap
curl -fsSL https://trwframework.com/install.sh | bash
01

Authenticate

Create an account or approve the device when prompted.

02

Choose clients

Select every supported surface used in this repository.

03

Write config

Review the repo-local files before opening your client.

Select clients during setup

The guided installer presents the current profiles below; the retired entry after them is listed for reference only and cannot be selected. Choose the surfaces already used by this repository; detailed file mappings belong in the configuration reference, not in the critical install path.

Installer profile preview

During setup, you can select one client or several. This preview is not interactive.

7 current profiles
  • Claude Code

    Hooks & agents

  • Cursor IDE

    IDE-native

  • Cursor CLI

    Headless-safe

  • OpenCode

    Command-first

  • Codex CLI

    Codex-native

  • GitHub Copilot CLI

    GitHub-native

  • Antigravity CLI

    Antigravity-native

Aider

The Aider target was retired because it never had a native adapter. Current installation rejects the identifier.

Retired; manual only

Not sure? Pick the client you will open immediately after installation. You can rerun the installer later without replacing your source code.

Compare profiles

Verification

Check
trw-mcp is on PATH
Command
which trw-mcp
What success looks like
Returns a path such as /home/you/.local/bin/trw-mcp
Check
.trw/config.yaml exists
Command
ls .trw/config.yaml
What success looks like
The repo-local config file exists
Check
Selected target_platforms were written
Command
grep -n "target_platforms" -A8 .trw/config.yaml
What success looks like
Shows every client surface you selected during install
Check
At least one client config exists
Command
ls .claude .cursor .codex .github .opencode .antigravitycli ANTIGRAVITY.md AGENTS.md opencode.json 2>/dev/null
What success looks like
One or more client-specific files or directories now exist in the repo

Run your first session

Do one small, real task. The first session should prove that TRW can start, preserve a useful checkpoint, record a discovery, and close with validation evidence.

  1. 01

    Start the session

    Loads prior learnings and recovers an active run if the work was interrupted.

    trw_session_start()
  2. 02

    Do the work

    Checkpoint whenever you cross a milestone so the session can recover instead of restart.

    trw_checkpoint("Milestone reached; tests next")
  3. 03

    Record a discovery

    If you learn something the next session should not rediscover, persist it while the context is fresh.

    trw_learn(
      summary="What changed",
      detail="Why it matters later"
    )
  4. 04

    Close the loop

    Run the project-native checks first, then record each actual command result before you persist the session state.

    npm test
    npm run type-check
    trw_build_check(command_results=[
      {"command_id": "tests", "label": "npm test",
       "command_class": "test", "exit_code": 0},
      {"command_id": "static_checks", "label": "npm run type-check",
       "command_class": "static", "exit_code": 0}
    ])
    trw_deliver()

If something feels off

Need deeper debugging? Head to troubleshooting for client-specific startup issues, install errors, and runtime recovery steps.

Hit a bug or have feedback? Run the /trw-feedback skill or call trw_submit_feedback to send it straight to the maintainer - PII is redacted before it leaves your machine.

When you need more control

Next

Quickstart gets TRW into the repo. Lifecycle explains what agents should do from research through delivery once the framework is live.