← ClaudeAtlas

decomposing-for-dispatchlisted

Use after designing a feature, when deciding whether its parts can be built in parallel - produces the contract file that `mnemo dispatch --contract` consumes
xyrlan/mnemo · ★ 2 · AI & Automation · score 66
Install: claude install-skill xyrlan/mnemo
# Decomposing a feature for dispatch Turn a designed feature into a **contract**: the pieces, and the boundary between them. Your input is the conversation you are already in — a brainstorm, a plan, or the maintainer describing the work out loud. Do not go looking for a plan file; plain plan mode writes none, and the contract is the durable artifact, not a copy of something else. Write the contract to `docs/superpowers/contracts/YYYY-MM-DD-<feature>.md`. Then stop. Dispatching is the maintainer's decision, not yours. Nothing reads that directory — `--contract` takes an explicit path, so the location is a filing convention and not a lookup. It is named here because it is where every contract in this repo actually lives, next to the specs and plans the decomposition came out of. An earlier version of this file prescribed `docs/mnemo/contracts/`, which no contract ever used and which has never existed in the repository. ## The test For every pair of candidate pieces, ask: > Can piece A be written and tested without reading the interior of B? If no, they are not two pieces. Merge them and ask again. Two pieces may depend on each other's **signatures** — that is what the contract records. They may not depend on each other's **internals**. ## `sequential` is a real answer If the work does not divide, write `verdict: sequential` and say why. That is a correct outcome, not a failure. A decomposition that always finds a cut produces only bad merges. Cutting by area — one p