loomwork
One loom, one status surface. An SDD lifecycle layer for superpowers: a status-frontmatter convention, a drift audit, a pre-merge close-out procedure, and a hook gate that enforces it.
Why
superpowers is a good execution engine: brainstorm, plan, execute. It does not
say what happens to specs and plans after the work ships. Left alone, specs
keep saying draft long after merge, and plans drift away from what was built.
loomwork adds that lifecycle as a plugin:
- One status-tracking surface. Spec frontmatter + acceptance checkboxes are the only "what's built" tracker. No second tool buys a sync engine — only a second source of truth that drifts.
- Plans freeze, specs live. A merged plan gets a DONE banner and stops changing; the spec stays current as the code evolves.
- Close out before merge. The spec and plan updates ride the same PR as the feature, so there is no follow-up PR after merge.
Full doctrine: references/PLAYBOOK.md.
Install
Claude Code
/plugin marketplace add kvasanova/toolshed
/plugin install loomwork
Requires the superpowers plugin (loomwork checks and tells you how to install it). Then, in each repo:
/loomwork:init
Codex
codex plugin marketplace add kvasanova/toolshed
codex plugin add loomwork@toolshed
Then, in each repo:
$loomwork:init
Codex skips plugin-bundled hooks until you review and trust them through
/hooks.
Initialization scaffolds the docs layout. It also upgrades a repo set up by an
older loomwork version: it removes the old doctrine block and the old Cursor
hooks, and moves the useful parts of an old roadmap file into a short product
context section. Idempotent. Non-default paths: create .loomwork.json (keys
specsDir, plansDir) before running init.
Upgrading to 0.5.0: re-run /loomwork:init once in each repo. Init never
commits; review its changes and commit them yourself.
What ships
| Piece | What it does |
|---|---|
loomwork:audit skill | Drift linter: closed issues on draft specs, merged PRs on unbannered plans, missing/stale verified dates. Exit 0/1/2. Not a CI gate by design. |
loomwork:close-out skill | Pre-merge procedure: freeze the plan and verify + update the spec — on the feature branch, same PR. |
loomwork:split-agents-md skill | Manual-only: splits one oversized agent guidance file into a tool-agnostic AGENTS.md plus thin host files (CLAUDE.md and friends) that import or reference it, with no duplicated prose. Moves any marker-guarded block into AGENTS.md whole, so init can still find and remove it. |
loomwork:model-assignment skill | Manual-only: right before execution, picks a model alias per task for subagent-driven-development implementer and reviewer dispatches. Held in the session by default; optionally written as a **Model:** line per task. No effect under executing-plans, which dispatches no subagents. |
| Hook gates (Claude Code + Codex) | Inject the close-out reminder when finishing-a-development-branch starts, and the loomwork doctrine at session start in opted-in repos. Codex observes explicit $superpowers:... prompts through UserPromptSubmit. |
/loomwork:init | Idempotent scaffolder, upgrader, and dependency preflight. |
Compatibility
Pinned upstream skill name this plugin's gate fires on:
finishing-a-development-branch (superpowers). The init preflight also checks
for brainstorming and writing-plans. Re-checked on upstream majors. A
renamed upstream skill makes the gate silently stop firing — if it goes quiet
after an upstream update, check this name first.
Support posture
No SLA. This plugin exists because its author runs it on their own repos; breakage from upstream majors is fixed when it bites those repos, not on report. Issues may go unanswered. Price that in before adopting.