MGM Garage Lab

Not a company. A garage for ideas ready to step outside.

Workflow Management Skill

Every new chat is a first date. Give it a diary.

AI assistants are sharp, but a fresh conversation starts without yesterday's project memory. Workflow Management gives the next session a durable handoff: what changed, what matters, what is blocked, and where to continue.

Three-panel comic labelled New Chat, No Context, Project Memory, showing an AI assistant moving from confusion to organized project memory with RECAP, TASKS and SESSION files.
New chatNo contextProject memory
Plain files

Markdown and JSON in a context folder you control.

Portable workflow

The project memory belongs to the work, not one chat window.

Beta path

Kiro adds an experimental team mode, still untested, while the stable package stays focused.

01 / The Problem

AI can write code. The hard part is remembering why the work looks like this.

Chat history is not a project plan. It is too long, too hidden, and too fragile. Workflow Management turns the living state of the project into plain files the assistant can read on purpose.

New session, same introduction.

You explain the repository, constraints, decisions, blockers, and next step again. The cost is not dramatic once. It becomes heavy because real work spans many sessions.

Lost handoffs

The useful ending of yesterday's work gets buried in a transcript instead of becoming today's starting point.

Context overflow

The assistant reads too much or too little because there is no compact, current source of truth.

Invisible decisions

Choices live in chat fragments instead of records you can inspect.

Forgotten blockers

Open issues return as surprises instead of known project state.

Messy restart

The first half hour becomes archaeology.

02 / Two Editions

One workflow contract, two practical packages.

The stable edition supports Claude Code and Codex. The Kiro edition is a separate beta package with the same personal workflow plus an optional team layer.

Claude / Codex edition

Stable

Open-source package for durable personal project context: sessions, RECAP, tasks, focus notes, meetings, roadmaps, decisions, and sprints.

  • Public repository: `workflow-management-skill`
  • Designed for file-based, local-first context
  • Tested with guided setup and session resume flows
Open repository
Same contract

Kiro edition

Public beta

A Kiro-specific package with the personal workflow skill and an optional team skill for shared roadmap, tasks, and assignments.

  • Public beta: personal skill tried, team mode experimental and untested
  • Team context stays separate from personal notes
  • Feedback wanted: try it on a scratch folder and open an issue
Open Kiro repository
It does not make the AI omniscient. It gives it a notebook, a map, and enough manners to read them before guessing.

Everything stays inspectable: Markdown records, JSON configuration, and normal project files under a context folder you control.

03 / Long Vision · Workflow Management

A hundred sessions.
Still one destination.

Weeks pass. Chats restart. Your project should still remember where it is going.

  1. 01 THE MAP / ROADMAP

    Set the course.

    Keep the long-term goal and the milestones that lead there.

  2. 02 THE LOGBOOK / SESSIONS

    Remember why.

    Carry decisions forward. Stop rediscovering yesterday.

  3. 03 THE WEATHER / RISKS

    Meet the detours.

    Record blockers and changes before they become surprises.

  4. 04 THE COMPASS / RECAP

    Find the next step.

    Start each session knowing what matters now.

  5. 05 THE VISION / PHASE GATES

    The destination.

    Check progress against the outcome you set out to reach.

04 / What It Keeps

The minimum memory a serious work session needs.

RECAP.md

Current state

Open work, blockers, and next steps without re-reading everything.

sessions/

Daily handoff

What happened today: decisions, progress, verification, and next actions.

tasks/

Work records

Goals, plans, progress, and done criteria.

focus/

Ideas and research

Notes that should survive the conversation.

roadmap/

Direction

Phases, milestones, sequencing, and decisions.

meetings/

Outcomes

Owners, unresolved questions, and follow-ups from real conversations.

05 / What Comes Next

The memory layer becomes the base for smarter project context.

Workflow Management is the first layer: durable state. Two later experiments explore how an assistant can choose the right documents instead of swallowing the whole project.

kb-graph A local-first knowledge graph over docs, decisions, tasks, and source references. Focus
graph-guided-rag A future RAG implementation where the graph narrows the search before semantic retrieval. Later
kiro-beta Colleagues try the personal skill and the experimental team mode; their feedback shapes the next version. Now