requirement-review
Challenge a feature or story before SDD in one of two modes. Mode --feature checks functional gaps and acceptance criteria; mode --tech grounds contracts and flows in the repository. Use when reviewing requirements or invoking requirement-review.
- License 许可
- MIT
- 最近更新
- 2026-11-08
- Stars
- ★ 6
- Forks
- 0
01 前言 Front Matter
- name
- requirement-review
- description
- Challenge a feature or story before SDD in one of two modes. Mode --feature checks functional gaps and acceptance criteria; mode --tech grounds contracts and flows in the repository. Use when reviewing requirements or invoking requirement-review.
02 资源文件 3
- 参考文档reference.md944 B
- 参考文档references/feature.md4.2 KB
- 参考文档references/tech.md4.5 KB
03 SKILL.md 正文
STOP - Read before ANY tool call
Read _shared/skill-boilerplate.md first. It holds the guardrails read, the session gate rule, the user sim rule, the one-step rule, the gate check, and the user chat language rule. The steps below are skill-specific only.
Writes (review report, Q&A history) are new artifacts: show the target path and summary, then wait for sim before any Write (guardrails.md section 2). Never write code.
Skill: requirement-review
Trigger
Negative signal: applies as defined in _shared/skill-boilerplate.md; the skill-specific signals are listed below.
Invoke when the user asks for: /requirement-review, requirement-review --feature, requirement-review --tech, "challenge this requirement", "review the acceptance criteria", or "check this story against the repository".
Not a trigger: a request to write a PRD (sdd-spec), to break steps into a checklist (split-story-checklist), or to change code (developer, *-developer).
Mode (mandatory, exactly one per run - no silent default):
| Mode | Explicit invoke examples | Playbook |
|---|---|---|
| --feature | --feature, feature, 1 |
references/feature.md |
| --tech | --tech, tech, 2 |
references/tech.md |
If the invocation names neither mode, or both: STOP after gate check (-1) and ask once, in the user chat language, then wait. Load no playbook until the mode is set. Do not assume a mode.
When invocation_context=orchestrated (a parent such as refine-story passes the mode), do not ask. A missing mode is blocked: return to the parent.
Question component form (ids fixed in English; labels rendered in the user chat language):
header: requirement-review mode
question: requirement-review mode?
single select:
- feature: functional review of the story or feature (invoke token --feature)
- tech: repository-grounded technical review (invoke token --tech)
Render the question and the two labels in the user chat language. Keep the mode ids feature and tech in English. The invoke tokens --feature and --tech map to them.
Outcome
One review report per run, with:
- Status token
READYorNEEDS_CLARIFICATION(_shared/sdd-artifacts/readiness-severity.md). - Open questions with severity, decision owner, and one answerable question each.
- Findings with stable ids (
refine-story/references/finding-format.md). - Atomized acceptance criteria (feature mode) or grounding and follow-ups (tech mode).
- Q&A entries appended to the Q&A history.
Flow outcome per _shared/skills-catalog/CATALOG.md (Flow terminal contract): done when READY and the handoff is named; open_question while any question waits for an answer; blocked when the source or the repository is missing.
Flow contract: reads, writes, where, when
- Reads:
STORY.mdorFEATURE.md,REFINE/qa-history.mdwhen present, or pasted text. - Writes: after
simonly:REFINE/requirement-review-feature.md(or-tech.md) andREFINE/qa-history.md. Pasted source: chat only. - Where:
features/NNN-slug/USnn/REFINE/; without a story folder,docs/backlog/<slug>-requirement-review-<mode>.md. - When to use: challenge a feature or story before SDD, with
--feature(acceptance criteria) or--tech(repository contracts).
Full flow: backlog refinement guide.
Lazy-load
| When | Path |
|---|---|
| Index (routing only) | {{TOOLKIT_ROOT}}/skills/requirement-review/reference.md |
| Mode playbook feature (only when mode = --feature) | {{TOOLKIT_ROOT}}/skills/requirement-review/references/feature.md |
| Mode playbook tech (only when mode = --tech) | {{TOOLKIT_ROOT}}/skills/requirement-review/references/tech.md |
| API contract standards (tech mode, only when an API contract is affected) | {{TOOLKIT_ROOT}}/skills/api-standards/SKILL.md, then one matching file under its references/ |
| Severity B / I / MINOR + READY gate | {{TOOLKIT_ROOT}}/skills/_shared/sdd-artifacts/readiness-severity.md |
| Clarify depth | {{TOOLKIT_ROOT}}/skills/_shared/backlog-item-types/clarify-depth.md |
| Finding types and block shape | {{TOOLKIT_ROOT}}/skills/refine-story/references/finding-format.md |
| Q&A history shape and flow rules | {{TOOLKIT_ROOT}}/skills/refine-story/references/qa-history.md |
| Storage paths and portable paths | {{TOOLKIT_ROOT}}/skills/_shared/sdd-artifacts/STORAGE.md |
Invocation contexts (direct vs orchestrated) |
{{TOOLKIT_ROOT}}/skills/_shared/sdd-artifacts/INVOCATION-CONTEXTS.md |
Selective retrieval (SR-NO-FULL-DUMP) |
{{TOOLKIT_ROOT}}/skills/_shared/sdd-artifacts/SELECTIVE-RETRIEVAL.md |
| Caveman Mode (if active) | {{TOOLKIT_ROOT}}/skills/_shared/caveman/CAVEMAN.md - Lite cap |
Never by default: do not preload references/feature.md and references/tech.md together; load only the selected mode. Do not load api-standards unless the tech playbook finds an affected API contract. Do not dump memory-bank/, a full PRD, or whole source files (SR-NO-FULL-DUMP).
Process
Step -1b - Caveman Mode (Lite cap)
- Read
{{SDD_ROOT}}/preferences.json(create{ "caveman_mode": false, "caveman_level": "full" }if missing). - If
caveman_modeis false: continue without compression. - If true: load
{{TOOLKIT_ROOT}}/skills/_shared/caveman/CAVEMAN.md; apply the Lite cap and prefscaveman_level; show once:[Caveman] Modo ativo (respostas compactas, level={effective}). Digite caveman off para desativar. - Honor
caveman on|off|status|lite|full|ultra(andstop caveman/normal mode) during the session. - Auto-Clarity and never-compress gates, drafts, and paths per
CAVEMAN.md.
0. Workspace and invocation context
Confirm the target repository only for --tech. Resolve invocation_context per INVOCATION-CONTEXTS.md: direct unless the parent handoff marks orchestrated.
0.5 Mode
Resolve --feature or --tech from the invocation or from the answer to the mode prompt. Then read only that playbook.
1. Locate source and prior history
- Source (one of):
features/NNN-slug/USnn/STORY.md(orTSnn),features/NNN-slug/FEATURE.md,REFINE/refine.mdbeside the story, or pasted text. Pasted text has no file: keep the report in chat until the operator confirms a write path. - Prior history first: read
features/NNN-slug/USnn/REFINE/qa-history.md(or the shortcutdocs/backlog/<slug>-qa-history.md) before asking anything. Do not re-ask a question whose entry isresolvedorsuperseded. - Never erase a decision. A changed decision is a new entry linked through
Supersedes/Superseded by, with the reason recorded in its Resolution field.
2. Run the mode playbook
Follow references/feature.md or references/tech.md. It produces findings, candidate questions, and mode-specific sections. Shared rules below apply to both modes.
3. Classify open questions
- Severity B, I, or MINOR per
readiness-severity.md. Depth perclarify-depth.md. - Decision owner: Product (value, scope, business rules, priority), Engineering (feasibility, contracts, data model, performance targets), or UX (flows, visible states, copy).
- Each question has one answerable question. Prefer a closed choice with stable ids (for example
confirm,adjust,cancel, or the domain ids listed in the question), labels rendered in the user chat language. Free text maps toadjustwhen it changes the proposal; otherwise the question stays open. Answer mapping:{{TOOLKIT_ROOT}}/skills/_shared/agents/QUESTIONS.md. - A product decision is never solved by assumption. Do not write a default into a criterion. Ask.
- Ask every open question in one batch at the end, including MINOR. Silence does not close a question. "I do not know" leaves it open.
Question entry (chat and report):
Q-003 [I] owner: Product
Question: <one answerable question>
Impact: <portable path>#<section>; CA-002.a
4. Atomize acceptance criteria
Applies to every criterion the source carries.
- Keep the parent id (
CA-002). When one criterion carries several independently observable outcomes, derive atomsCA-002.a,CA-002.b, and so on. - Never renumber an existing id. A new atom takes the next free suffix under its parent.
- Each atom has Given / When / Then with an observable Then. Include the boundary atom (empty, minimum, maximum, duplicate, where the domain has them) and the failure atom (what the user sees and what is persisted).
- An atom must be checkable by someone who did not write it. Reject "works correctly", "as expected", "properly", and "funciona corretamente" as Then.
5. Report and Q&A
Output shape (report body; prose follows the user chat language, tokens stay English):
# Requirement review - <title>
| Field | Value |
|-------|-------|
| **Mode** | --feature \| --tech |
| **Invocation context** | direct \| orchestrated |
| **Source** | <portable path or "(pasted text)"> |
| **Updated** | YYYY-MM-DD |
Status: READY | NEEDS_CLARIFICATION
## Open questions
| ID | Severity | Owner | Question |
|----|----------|-------|----------|
| Q-003 | I | Product | <one answerable question> |
(Open items only. Resolved and superseded items live in the Q&A history.)
At every `open_question` stop and at close, you must show this full table in chat with every open question row (ID, Severity, Owner, Question), localized per `_shared/sdd-artifacts/LIVE-STAGE-TABLE.md`. Never show only the changed row or summarize rows. A question the operator has resolved leaves the table and stays only in the Q&A history.
## Findings
[FND-001 | I | GAP] <section heading>
Evidence: <portable path> | no-evidence
<what is ambiguous, missing, contradictory, risky, or dependent>
Recommendation: <recommendation, not a decision>
## <mode section>
<feature: atomized criteria table; tech: grounding and follow-ups, see the playbook>
## Next
<handoff line from step 6>
- Finding ids, types, and block shape:
refine-story/references/finding-format.md. Do not invent a second severity vocabulary. - Q&A entries: append new questions with
Status: openbefore waiting for answers, using theqa-history.mdshape. SetOrigintorequirement-review --featureor--tech. Keep question ids stable; allocate the next freeQ-andD-ids. - Persistence (after
sim):features/NNN-slug/USnn/REFINE/requirement-review-feature.mdorrequirement-review-tech.md, with the Q&A history atREFINE/qa-history.md. Shortcut when no story folder exists:docs/backlog/<slug>-requirement-review-<mode>.mdwithdocs/backlog/<slug>-qa-history.md. Source is a FEATURE.md with no story folder: report in chat and persist only to a story folder the operator names, or the shortcut. Do not invent a feature-rootREFINE/folder. - Portable paths only in the file, the handoff, and the history. No OS absolute paths.
6. Status and handoff
Status follows readiness-severity.md: READY only with no unanswered B, I, or MINOR question. Recompute it on every refresh.
| Status | Handoff |
|---|---|
READY, mode --feature, no repository involved |
sdd-spec - <portable-story-path> |
READY, mode --feature, story touches a repository |
requirement-review --tech - <portable-story-path> next, then sdd-spec |
READY, mode --tech |
sdd-spec - <portable-story-path>, with the listed technical follow-ups |
NEEDS_CLARIFICATION |
Answer every open question, cite the Q&A history path, re-run the same mode. Do not claim the item is ready for sdd-spec |
Technical follow-ups are not questions. They never block READY; they travel to sdd-spec as a list.
Must not
- Decide a product question by assumption, or write a default into an acceptance criterion
- Claim READY while any unanswered B, I, or MINOR question remains
- Write code, edit application files, or write a PRD or PLAN
- Assert a repository fact without a search hit (
--tech); recordnot-foundinstead of guessing - Copy the
api-standardsrules into the report; cite the reference file - Load both mode playbooks, or load the api-standards pack without an affected API contract
- Erase or rewrite a prior answer, resolution, or impact; supersede it with a linked entry
- Put more than one question in one entry
- Call an external tracker, work-item API, or MCP work-item tool
- Write OS absolute paths or InstallRoot embeds into the report, history, or handoff
- Describe the design by reference to any external product, repository, or skill; state behavior in toolkit terms
Handoff examples
requirement-review --feature - features/004-export/US01/STORY.md
requirement-review --tech - features/004-export/TS02/STORY.md
/sdd-spec - features/004-export/US01/STORY.md
Cite REFINE/qa-history.md in the typed handoff when present. Do not paste the history into the slash line.
04 同仓库更多 Skills
user-own-skill
Operator-owned skill that is not part of the toolkit.
vue-developer
Implement or fix small-to-medium Vue 3 features without full SDD (Composition API, Pinia, configured project test runner; Vitest is a fallback for new projects). Use for isolated Vue work or when invoking /vue-developer.
test-coverage
Run .NET Coverlet coverage, report Sonar-aligned metrics, and evaluate against a threshold (default 80%). Use for coverage reports or when invoking /test-coverage.
split-story-checklist
Break refined backlog steps into a dependency-aware task checklist (backend, frontend, tests). Prefer features/... story folders. Use when breaking down tasks or invoking /split-story-checklist.
sdd-spec
Create a PRD for a new feature or change (agent PRD .md; language per LANGUAGE.md + preferences under features/). Feeds sdd-plan. Use when creating a spec or invoking /sdd-spec.
sdd-plan
Create a baby-step PLAN from an existing PRD (agent PLAN .md; language per LANGUAGE.md + preferences). Feeds sdd-develop. Use when creating a plan or invoking /sdd-plan.