ProAgents Workspace (paw)

The convention-first workspace for AI coding agents — a lightweight developer CLI that makes existing AI coding agents work better inside a predictable, inspectable, convention-driven environment.

npm install -g @reposell/proagents-workspace

cd my-project
paw init README.md   # Workspace-aware repo — intent inferred from YOUR README, persisted to .paw/workspace.yaml
paw status           # where am I, what is here, what is possible
paw verify           # runs detected lint/typecheck/test/build — no configuration needed
codex                # your existing agent keeps working

No CLI at all? Paste this into any coding agent (Claude Code, Codex, Gemini CLI, …):

Setup https://github.com/EnzoVezzaro/proagents-workspace for @README.md (project's instructions).

The agent follows the bootstrap protocol and runs npx @reposell/proagents-workspace@latest init README.md — the six-stage bootstrap: resolve → acc → shield → proagents → reposell → paw (install is a stable alias).

For the complete install → set up → complete contract — a ready-to-paste start prompt, the long-run operating playbook (finish line, stop conditions, evidence-checked subagents, self-review), the verification gates, failure handling, and the final report — see the PAW Installation Runbook.

No runtime required. No cloud account required. No desktop application required.

Works with Codex, Claude Code, OpenCode, Gemini CLI, DeepSeek Harness, and custom agents — the Workspace improves the environment around your agent; it never replaces it.

Commands

| Command | What it does | |---|---| | paw init [file] | The workspace BOOTSTRAPPER: runs the six-stage bootstrap (resolve → acc → shield → proagents → reposell → paw), writing each layer's configuration; with a file (e.g. README.md) the inferred intent is persisted as the project: block | | paw install [file] | Stable alias of paw init — the same six-stage bootstrap, both spellings everywhere | | paw lifecycle show | The canonical phase flow this workspace runs (create → … → operate) | | paw status | Where am I, what project, what branch, what agents, what verification | | paw research | Product/environment discovery (ACC-aware); hands agent requirements to ProAgents | | paw doctor | Diagnostics — optional components are marked optional, never failed | | paw verify | Runs detected or configured checks (lint → typecheck → test → build) | | paw diff | Git working-tree diff | | paw agent list | Detected agents with honest integration tiers | | paw --json | Every important command has machine-readable output |

Principles

• Convention first — everything that can be inferred locally is inferred before asking you to configure it
• Local first — the default workspace operates directly on your current repository; nothing is cloned or moved
• Progressive — isolated runtimes (Docker/E2B), lifecycle orchestration, and the desktop UI are opt-ins, invisible until they provide value

Links

• Documentation
• PAW Installation Runbook — the complete install-to-complete contract for an AI agent
• Changelog
• Contributing / source

MIT © Enzo Vezzaro