clawplays/ospec

npm install -g @clawplays/ospec-cli (official package @clawplays/ospec-cli, command ospec; verify with ospec --help)

Usage

• •ospec init . — initialize OSpec in a project directory (options: --summary, --tech-stack, --architecture, --document-language en-US/zh-CN/ja-JP/ar)
• •ospec change <change-name> — create and advance one change for a requirement, doc update, or bug fix
• •ospec goal <goal-name> --target codex --execution-model controller — full goal workflow with parallel worker dispatch and reviewer gates
• •ospec verify changes/active/<change-name> then ospec finalize changes/active/<change-name> — verify and archive an accepted change
• •ospec session hook --target claude --apply — installs Claude Code hooks that announce dispatches and hard-block subagent dispatch while a required decision is pending

Key features

• ✓Spec-driven work saved to the repo (.ospec/): proposal, design, implementation plan, tasks, reviews, and verification evidence files
• ✓ospec change fast flow (init → change → verify/finalize) plus ospec goal full workflow with task-graph controller, native-subagent dispatch, and durable evidence
• ✓Agent-agnostic: any assistant (Codex/GPT, Claude Code, Gemini, OpenCode, or plain CLI) can resume where the last one stopped
• ✓Loop Engineering: token-bounded loop commands (ospec loop), parallel conflict-safe worker batches, reviewer gates, and repair convergence guards
• ✓Plugins for UI review (Stitch) and runtime validation (Checkpoint); skill management for Codex and Claude Code

Best for

Repo-resident, spec-driven development workflow that makes AI coding change context inspectable, reviewable, and resumable across sessions and different agents.

Caveats

• ⚠Requires Node.js 18+ and npm 8+
• ⚠Force/auto archive is never automatic — it requires the force flag, an exact change-name confirmation, and an audit reason
• ⚠Native subagent dispatch requires harness support; OSpec never starts Orca/Codex/Claude as a fallback and blocks if no supported harness reports capability
• ⚠This is a CLI workflow framework, not a runnable MCP server (it targets MCP-based agents as clients)