Getting started

Install ccm and the plugin, fire your first orchestration, and learn the handful of commands you will actually type.

cc-master has two installable pieces: ccm, the engine CLI that owns all state, and the plugin, which teaches your agent harness to orchestrate. One installer sets up both. ccm is a hard prerequisite — without it the plugin refuses to start an orchestration — so the installer always places it first.

Install

# latest of both version lines (plugin + ccm)
curl -fsSL https://raw.githubusercontent.com/nemori-ai/cc-master/main/install.sh | bash

# pin either line independently — the two release on separate tracks
curl -fsSL https://raw.githubusercontent.com/nemori-ai/cc-master/main/install.sh | bash -s -- \
  --ccm-version ccm-v0.21.0 --plugin-version v0.20.1

# target one harness, or fan out to every supported harness on the machine
curl -fsSL https://raw.githubusercontent.com/nemori-ai/cc-master/main/install.sh | bash -s -- --harness claude-code
curl -fsSL https://raw.githubusercontent.com/nemori-ai/cc-master/main/install.sh | bash -s -- --all-harnesses

The installer detects your OS/arch, downloads the right ccm binary, verifies every downloaded asset against the release’s SHA256SUMS, then installs the adapter for each detected harness. A checksum mismatch stops the install — treat it as a release-integrity failure, not a prompt to bypass.

Requirements: Node.js 22+, unzip, and a SHA256 tool (sha256sum, shasum, or openssl); online installs also need curl or wget. Claude Code installation uses the claude CLI (≥ v2.1.195). Supported harnesses: Claude Code, Codex, Cursor, kimi-code. ccm ships for Linux and macOS (x64/arm64); Windows is not supported yet.

Verify the install

ccm --version

Then start a session in your harness and use its entrypoint:

Add --resume to any of them to take over an existing board instead of starting fresh.

Your first orchestration

Give it a goal that has real shape — one shared foundation, then independent parallel work:

/cc-master:as-master-orchestrator Internationalize the app to 6 locales
  (i18n framework + per-locale translation + locale routing)

Here is what happens after you hit enter:

1. Bootstrap. The entrypoint fires the bootstrap hook, which creates a board — one JSON file under ~/.cc_master/boards/ that becomes the single source of truth for this run.
2. Goal Contract. Your sentence is treated as evidence, not as the plan. The orchestrator rewrites it into a short, testable Goal Contract and asks you only about ambiguities that would change the outcome — then confirms it before any task exists.
3. DAG. The goal is sliced into a dependency graph: extract strings and wire the framework first, then six locale tasks that can all run at once.
4. Parallel dispatch. Ready tasks go to background workers immediately — the groundwork may get a stronger model tier, the mechanical translations a cheaper one.
5. Decision package. When a call is genuinely yours (“translate product terms or keep them in English?”), it surfaces with context and options — while everything that doesn’t depend on the answer keeps running.
6. Endpoint verification. A green gate or a worker’s self-report never counts as done. The orchestrator verifies each result independently before marking it done.
7. Stop. /cc-master:stop runs a completion check against the Goal Contract, then archives the board. You can --resume it later — even from a different session or harness.

The everyday five

The in-session commands are harness-specific; ccm commands always run in your terminal.

• Status — ccm status-report show: progress, blockers, critical path, next actions.
• Watch — ccm web-viewer open: the live plan as a read-only graph in your browser.
• Answer a waiting decision — /cc-master:discuss <decision> (Codex: $cc-master-discuss, Cursor: /discuss, kimi-code: cc-master:discuss).
• Stop — /cc-master:stop (Codex: $cc-master-stop, Cursor: /cc-master-stop, kimi-code: cc-master:stop). Archives the board; resumable later.
• Resume — add --resume to your harness entrypoint; the new session reconciles live evidence and picks up from the breakpoint.

Honest limits

Not every harness gets every capability. kimi-code ships the skills, commands, and core hooks but has no custom subagent roles, no Workflow equivalent, and no CLI quota signal; Codex and Cursor never auto-switch accounts; Cursor paces against its billing period. The Feature Manual (linked below) is the honest current/partial/target boundary — check it before assuming a capability exists on your harness.