← ClaudeAtlas

keel-discoverlisted

Use when starting any feature, change, or project whose requirements are not yet nailed down — turns a vague idea into a user-approved, evidence-grounded spec before any code is written, including a prior-art scan of what already exists outside the repo. Also the re-entry point when shipped work's Signals came back negative — the requirement was wrong, not the code. Stage 1 of the keel pipeline; hands off to keel-plan.
AWenSu/keel · ★ 0 · AI & Automation · score 75
Install: claude install-skill AWenSu/keel
# keel-discover — Requirements Discovery & Spec ``` INPUT a raw idea or request, plus access to the real codebase OUTPUT a user-approved spec: scope locked, test seams agreed, every requirement grounded in evidence rather than recollection ``` **Arriving from a negative signal?** When the entry point is shipped work whose `## Signals` came back wrong, the first question is not "what should we build" but **"which part of the original requirement was mistaken"** — quote the plan's Signals line and the observation that contradicted it before re-running the intake below. A requirement that failed in reality has evidence a fresh one does not; wasting it is how the same wrong thing gets specified twice. Missing INPUT → `BLOCKED: 缺 <idea | codebase access> → 退回 the requester`. This stage is the pipeline's entry point, so there is no earlier stage to route to: an idea too vague to state at all goes back to the person who asked, and no codebase access means the evidence rule below cannot be met. Turn "I want X" into a written, approved spec. Nothing downstream (plan, review, execution) can be better than the spec it started from. <HARD-GATE> Do NOT invoke any implementation skill, write any code, scaffold any project, or take any implementation action until a design has been presented and the user has approved it. This gate has no exceptions for "simple" projects — simple projects are where unexamined assumptions waste the most work. </HARD-GATE> ## When to skip Skip