Weft

Weft (Claude Code plugin)

Spec-driven AI development orchestration, woven on jj (the loom) and beads (the warp/brain). This plugin packages Weft's orchestration prompts as native Claude Code skills and agents.

Skills

• /weft:execute [epic-id] — weave one wave of ready picks: form the shed, isolate per-pick workspaces, dispatch executor agents, integrate in dependency order, resolve conflicts, and land conflict-free picks. Repeats until the epic's ready set is empty.
• /weft:new-project [project description] — plan a new project into the warp: adaptive questioning, parallel research, and requirement extraction, then emit a warp-plan.json for human approval.
• /weft:discuss [epic-id] — shape HOW a phase gets built: adaptive gray-area questions whose locked decisions land in the epic's bead design field for the planner to consume.
• /weft:verify-work [epic-id] — interactive UAT: walk the epic's deliverables y/n one at a time, diagnose failures, and file fix picks under the epic.

Agents

Dispatched by the skills via the Task tool: weft-executor (per-pick TDD), weft-planner (warp generation), weft-reviewer (per-pick review), and weft-resolver (conflict resolution).

Prerequisites

The skills drive these CLIs, which must be on PATH:

• weft — the engine binary (sits next to bd and jj).
• bd (beads) — the dependency graph and task state (the warp/brain).
• jj (jujutsu) — the colocated VCS (the loom).
• gh — GitHub operations.

Installation

Install via the engine, which pins the plugin to the running binary's release:

weft install

weft install registers this repository as a Claude Code plugin marketplace and installs the weft plugin (weft install --uninstall removes it). See https://github.com/seanb4t/weft for details.