← ClaudeAtlas

dslisted

The spec loop for a repository whose product model lives in `.ds/` — look up what a feature is, where it lives in the code and what it touches; implement inside that scope; then reconcile the model with what was built. Use whenever this repo has a `.ds/` directory, when CLAUDE.md carries a `<!-- ds: … -->` stamp, or when the user mentions specs, the product model, drift, or the code map.
tuna781/dspec · ★ 1 · AI & Automation · score 62
Install: claude install-skill tuna781/dspec
# Working against a dspec model This repository's product model lives in `.ds/`, written in **dspec-lang**, and **the model is the source of truth** for what the product does, where each feature lives, and what depends on what. `CLAUDE.md` is a compiled pointer at it — never edit that file by hand; it is overwritten. Everything is local. There is no server, no token, no network call. The CLI reads `.ds/` and your actual source files. ⚠️ **dspec is a toolkit; you are the brain.** Every command either measures something readable from the checkout or writes something mechanical. None of them decides what a feature is, which files deserve one, or whether a description is still true — those are judgements, and they are yours and the user's. Where the tool reports and does not act, that is deliberate: act on the report. ## Start here **`.ds/index.md` is the entry point.** One read gives you every feature, what it is, and which files it lives in. Read it before you read any source. ``` .ds/product.md what this product is, and the rules every change must respect .ds/glossary.md what the words mean .ds/index.md GENERATED — every feature, one line .ds/features/*.md one file per feature — the only prose anyone writes ``` The path of a feature file **carries no meaning**: subfolders under `features/` are a convenience and are ignored. `area:` is the only grouping. ## The loop **Code first, then reconcile.** The model is written *after* the code exists, so `.ds/