moai-kanban-foreman

Dispatches picked kanban backlog cards to isolated worktree workers and verifies completion evidence.

Updated Aug 29, 2026
One-click install
npx skills add https://github.com/Seung-zedd/secure-file-upload --skill moai-kanban-foreman-seung-zedd
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: moai-kanban-foreman
Source: https://github.com/Seung-zedd/secure-file-upload/tree/main/.claude/skills/moai-kanban-foreman
Command: npx skills add https://github.com/Seung-zedd/secure-file-upload --skill moai-kanban-foreman-seung-zedd

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Running an unattended backlog loop safely requires strict boundaries: the agent must never pick work itself, never answer approval gates on the operator's behalf, and never trust a worker's claim of completion without reading evidence. This Skill encodes one idempotent foreman iteration that enforces all of those rules mechanically. ## Core Features & Use Cases - Queue watch and dispatch: Monitors the backlog file via a checksum-polling watcher, reads the queue with moai todo list --json, and dispatches only the oldest operator-picked card to a single worktree-isolated worker. - Evidence-based completion: Advances a card with moai todo done only after reading the worker's evidence file containing verbatim passing check output; missing or stale evidence leaves the card blocked. - Hard unattended boundaries: Removes AskUserQuestion from the tool pool, forbids pushes, merges, PRs, and concurrent write-capable workers, and reports blocked cards with the decisions they wait on. - Use Case: An operator starts a session, runs bare /loop, and backgrounds it; each iteration the foreman watches the queue, dispatches one picked card to a worktree worker, collects its evidence, and writes a two-to-six-line report the operator reads on reattach. ## Quick Start Run a bare /loop in the project session and let the foreman watch the backlog, dispatch the next picked card, and report what evidence it read.

Frequently Asked Questions about moai-kanban-foreman

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run an unattended kanban backlog loop with an AI agent?▼

Start a session in the project, run a bare /loop, then background the session so loop tasks keep running without a terminal. Each iteration watches the backlog queue, dispatches one picked card to an isolated worker, and reports what it read.

How does the foreman verify a worker actually finished a task?▼

Completion is read, never trusted: the foreman advances a card only after reading the worker's evidence file containing verbatim passing check output and named gaps. A missing, unreadable, or stale evidence file leaves the card picked and blocked.

Can the foreman pick or reorder backlog items itself?▼

No. Only items whose state is already picked are dispatchable; the foreman never runs moai todo add or moai todo next, and never invents, rewords, or reorders cards. An empty queue is treated as a legitimate idle state.

What happens when a card needs a human decision mid-loop?▼

The foreman never answers approval gates on the operator's behalf. The card stays picked, is named blocked-for-operator in the report along with the decision it waits on, and the iteration moves on.

Does the loop push branches or create pull requests?▼

No integration actions are allowed: no push, pull request, merge, branch deletion, or worktree disposal. The card's branch stays unpushed and its worktree remains until the operator integrates them manually.

Why does the queue watcher poll with checksums instead of file events?▼

The backlog file is replaced atomically on every mutation, so tools following a single file handle miss changes. A checksum poll every five seconds emits one line per change at minimal cost and re-triggers the idempotent iteration.