菜单
KNOWLEDGE (data) — the 19-item agent-spec validation checklist. The item definitions the validate-spec / todoify procedures RUN against. No sequence of its own.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/nWave-ai/nWave-experimental --skill nw-ab-validation-checklist命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Thin prompt-level router for explicitly authorized Auto M/L work: reuse the acceptance-designer, paradigm crafter, independent examiner, and Git evidence without creating another controller.
Cross-cutting normative invariants — lean public routing core for global gate/construction doctrine and on-demand knowledge lenses. Cite clause ids; never re-declare.
Establishes durable architecture, reuse, boundaries, cross-layer algebra, residual stress behavior, paradigm, and prefactoring decisions for later DeliveryContract compilation.
基于 SOC 职业分类
正在显示 SKILL.md
| name | nw-ab-validation-checklist |
| description | KNOWLEDGE (data) — the 19-item agent-spec validation checklist. The item definitions the validate-spec / todoify procedures RUN against. No sequence of its own. |
| user-invocable | false |
Kind: KNOWLEDGE (data). The canonical 19-item list. nw-ab-validate-spec runs all 19 as ordered tasks; nw-ab-todoify-file runs items #14 + #15. This skill holds the definitions (SSOT), not the run-sequence.
| # | Item | Gate |
|---|---|---|
| 1 | Frontmatter | name, description, model, tools, maxTurns, skills present |
| 2 | Line Count | wc -l under 400; domain knowledge in skills |
| 3 | Divergence Only | zero instructions restating Claude defaults |
| 4 | Language Tone | zero CRITICAL/MANDATORY/ABSOLUTE outside skill loading |
| 5 | Examples | 3-5 ### Example sections |
| 6 | Least Privilege | tools list minimal; no Write/Edit for reviewers |
| 7 | Safety by construction (GDP-0) | tool-surface frontmatter + typed grammar first, not prose; a hook only last-resort with a recorded reason |
| 8 | Affirmative Phrasing | zero negatively-phrased rules |
| 9 | Terminology | one term per concept |
| 10 | Description Quality | description states WHEN to delegate |
| 11 | Skill Loading | mandatory section, imperative language, explicit ~/.claude/skills/nw-{name}/SKILL.md |
| 12 | Load Directives | every phase has Load: matching frontmatter skills |
| 13 | No Orphan Skills | every frontmatter skill appears in a phase |
| 14 | Workflow Format | numbered task list (N. **Name** — action. Gate: condition.), not prose |
| 15 | Success Criteria Format | numbered/checkbox list, not prose |
| 16 | Caveman House Style | dry/declarative, tables + compact lists, lean body |
| 17 | Reasoning Mandate Present | ## Reasoning Mandate section present |
| 18 | A05/A06 Literal Anchors | verbatim You MUST load your skill files OR Your FIRST action before any other work AND ~/.claude/skills/nw- |
| 19 | One Job, One Trigger | one job + one trigger; KNOWLEDGE vs PROCEDURE classified; PROCEDURE states trigger + sequence + composition; zero multi-job parameterized switches; existing command + agent NAMES preserved (no rename/proliferation; new command is the rare /nw-* exception) |
