catblob-hash

Mework — Development skill for Claude Code

Development community

Mework - local-first agent workbench for Windows (Mew.

How to install Mework

This entry records only its repository, not the path inside it, so there is no exact command to give. Open catblob-hash/Mework and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Mework does

Mework - local-first agent workbench for Windows (Mew. Work.)

Alternatives in Development

IPolloWork — Enterprise-grade, local-first Agent Workbench for people and agent teams 4.9k ★
Alethe Agents — A local-first desktop workspace for running, organizing, and resuming multiple coding agents and shells with r 511 ★
Lacp — Local-first policy, evidence, memory, and recovery controls for Claude, Codex, Hermes, and other CLI coding ag 300 ★

README

Mework

**Mework — Mew. Work.**

**English** | [简体中文](README.zh-CN.md)

📖 **Documentation:** [catblob-hash.github.io/Mework](https://catblob-hash.github.io/Mework/en/index.html) — how to work with Mework, the tool-description file (prompt profile) reference, and how to configure skills, MCP servers and hooks.

Mework is a local-first agent workbench for Windows. It pairs a React 19 interface with a Rust host built on Tauri 2: the host owns persistence, capability discovery, tool execution, and every safety boundary, while a bundled AI SDK sidecar is the only path to model providers. You bring your own API keys; nothing leaves your machine except the model calls you configure.

Highlights

**A formally specified agent kernel.** The turn/round/tool/subagent lifecycle is specified before it is implemented: a CSP-M interaction protocol ([`formal/csp/AgentKernel.csp`](formal/csp/AgentKernel.csp), the highest authority) and a TLA+ safety model ([`formal/tla/AgentKernel.tla`](formal/tla/AgentKernel.tla)) are checked with ProB — model checking, CSP refinement with 93 scenario assertions (21 accepted, 72 rejected), guard-deletion mutation matrices (63 TLA + 92 CSP), and bidirectional trace replay against the dependency-light Rust kernel crate (`src-tauri/agent-kernel`). At runtime a shadow kernel observes the real host event stream and reports any divergence from the spec.

**Bring your own providers.** OpenAI Responses, OpenAI Chat Completions, Anthropic Messages, Google, Azure OpenAI, Amazon Bedrock, Google Vertex, xAI, and generic OpenAI-compatible endpoints — all through an embedded [Vercel AI SDK](https://sdk.vercel.ai) sidecar compiled to a single executable. Providers are user-defined; there is no built-in vendor list to outgrow. Two families step outside the plain "base URL plus key" shape. **OpenAI Codex**: sign in with your ChatGPT subscription (Plus / Pro / Team) instead of an API key — the OAuth flow runs in the host, tokens stay encrypted on this machine