omp packages

@direal/miaw0.4.2

MiAW artifact-backed workflow CLI and local server

jsnzwu Published 19 days ago MIT ~342 downloads/mo

Untested on omp

This is a pi package. No one has reported whether it works on omp yet. If you try it, add your result below.

Did it work for you?

Result

One report per person. A new report replaces your old one.

Reports 0

No reports yet.

Readme

MiAW

Minimal Agentic Workflow — project-local state and evidence for work done with coding Agents.

MiAW gives an Agent a recorded task to resume, rather than asking it to reconstruct the project from chat history. It works alongside your coding Agent and Git; it is not a model, an IDE or a hosted execution service.

中文

The design

Human decisions
      ↓
Coding Agent + skills ── aw CLI ── project records (.aw/)
                            ├── Agent Server → delegated workers
                            └── Project Index → source retrieval

The host Agent reasons and uses tools. Skills explain the working procedure. The CLI validates records and state transitions. Agent Server runs bounded delegated work; Project Index helps locate source evidence. Neither a worker report nor a search result counts as acceptance by itself.

Record Question it answers
Project and Blueprint What are we building, and which design rules apply?
Roadmap Which candidate work should come next?
Task What was requested, planned, changed and checked?
Agent evidence What did each delegated worker actually do?
Knowledge Which accepted lessons should later tasks retain?

A task follows a recorded work loop:

Intention → Analyze → Plan → Execute → Verify → Commit → Archive

Brainstorm, Review and Consolidate are used when needed; skipping an optional step requires a recorded reason. Blocked work stays resumable by task ID. Verification, risk acceptance and commit approval are separate decisions: an Agent saying “done” does not close the task.

Durable design and execution records live in .aw/; Git records source changes. Disposable service state and index caches live in .aw/.runtime/ and stay out of Git and synchronization. Worktrees share a worker pool only for the same workflow owner on the same host platform; WSL and native Windows have separate runtime pools.

Start

Requires Node.js 18+ and a supported host: Linux x64 with glibc 2.28+, Windows x64 or macOS Apple Silicon. npm installs the platform binary; Rust is not required. Use Git for commits and isolated worktrees, or configure vcs = "none" for a project without Git.

npm install -g @direal/miaw
aw --version
cd your-project
aw init "Short project summary"
aw install --target agents
aw status

agents installs skills into .agents/skills/. Use --target claude-code for Claude Code or --target codex for hosts that use .codex/skills/. Pi users can also install the optional status integration with pi install npm:@direal/miaw.

Open your authenticated coding Agent in that project and use the skills through its supported interface. For example:

$aw-init Establish the project goals and constraints
$aw-task Add rate limiting to the API
$aw-resume <task-id>

Initialization and record inspection need no model account. Delegated execution additionally needs a configured provider and passing readiness checks. Linux/WSL workers require Bubblewrap; native Windows execution requires explicit unsandboxed opt-in. Roles and permitted write paths are not a substitute for OS isolation.

Inspect and maintain

  • aw status shows task state; aw tui provides an interactive view.
  • aw docs getting-started --lang en explains provider setup and host integration. Use --lang zh for Chinese; the manual ships with the package and works outside a source checkout.
  • aw update status inspects the installed release; aw update asks for authorization before the first update. Update each host separately.
  • If aw --version still shows an older build, inspect command precedence with type -a aw on Linux/WSL or Get-Command aw -All in PowerShell. A Cargo installation can precede the npm command.

Manage workflow state through the CLI or skills, not by editing task JSON. Review prompts, responses and logs before sharing .aw/: durable evidence can contain private project information.

MIT licensed. Model accounts and provider credentials remain under your control.