# Specify

> Write the grounded spec — EARS requirements, user flows, and Gherkin acceptance scenarios. Output — requirements.md + <flow>.feature files.

- Skill: `nvzver/specify` (Agent Skill)
- Install (CLI): `npx skillmds@latest add nvzver/specify`
- Raw SKILL.md: https://api.skillmd.com/api/skills/nvzver/specify/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: NVZver (https://skillmd.com/u/nvzver)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/nvzver/specify

---


> **Trace.** On load, print first: `=============== [lsa/skills/specify/SKILL.md] [lsa] ===============`

# LSA Specify

See [CORE.md](../../CORE.md) §5 (standards) and §8 (templates).

## Role

Spec author.

## Goal

A grounded, technology-agnostic spec: EARS requirements, user flows, and one Gherkin scenario set per flow.

## Inputs

| Input | Source |
|-------|--------|
| Intent + cited codebase facts | `discover` |

## Steps

1. For each user flow, answer Flow / Success / I/O / Test (CORE §3). (→ user flows)
2. Draft EARS requirements — "While/when … the system shall …" (CORE §5). No mechanism (CORE §7). (→ draft requirements)
3. Draft one Gherkin `.feature` per flow; each `Given` grounded in a fact from `discover` (CORE §8). (→ draft scenarios)
4. Show the full draft spec (per [`core/output`](../../../core/skills/output/SKILL.md) Rule 7 *Delivery test*); take human approval; **then** write `requirements.md` + the `<flow>.feature` files — show → approve → write per Rule 7 *Authorization boundary*. (→ approved spec on disk)

## Output

`${specs_root}/features/<name>/requirements.md` + `<flow>.feature` files — written only after the Step 4 approval; nothing on disk before it.

## Constraints

- Behavior, not mechanism — libraries and APIs are the implementer's choice.
- Every `Given` cites a real fact from `discover`; an ungrounded `Given` is `[ASSUMPTION]`.

---

`/lsa:specify` — manual invocation.