camerata
v0.1.0
Published
Camerata — a multi-agent orchestration engine and playbooks for Claude Code and Codex CLI orchestrators. Full release in development.
Maintainers
Readme
camerata
Camerata — a multi-agent orchestration engine and playbooks for Claude Code and OpenAI Codex CLI orchestrators.
A strong planning model owns the goal, decomposition, and judgment; bounded worker agents execute in isolated git worktrees; nothing merges on a worker's self-report. Successor to a private harness proven in consulting use.
Full release in development.
Install
One repo, three paths — all on the same engine version.
# Claude Code
/plugin marketplace add dvause/camerata
/plugin install camerata@camerata
# Codex CLI — installs the skills, registers the MCP server
npx -y camerata setup-codex
# bare CLI
npm i -g cameratacamerata doctor reports what the current machine supports: git, each backend
CLI, and codex sandbox availability. On native Windows the codex backend
refuses and the claude backend carries the run.
Playbooks
Four skills drive the engine. Each is portable SKILL.md — the same file works
in Claude Code and Codex.
| skill | stage | produces |
| --- | --- | --- |
| camerata-spec | fuzzy ask → testable spec (solo) | approved spec doc |
| camerata-plan | spec → dependency-ordered runs (solo) | orchestration plan |
| camerata-build | plan → parallel workers, review→fix loop | integration/<project> branch |
| camerata-audit | read-only auditors, findings bus | ranked report |
Every stage boundary is a human gate. Workers never run git, nothing merges on a worker's self-report, and a run is not over until the close gate exits clean.
License: Apache-2.0
