2026 · 0905
日期:2026-09-05 · 类型:版本发布(首版)· 关联版本:v1.0.0
一套可分行业复用的 AI Harness 规约体系首次发布——5 大域、26 个场景,每个场景给出「使用规约细节 + 可复制的提示词」。建立在 AGENTS.md + SKILL.md + MCP 三件套之上,不自造 DSL、不绑定任何厂商。
1. 发布内容
1.1 规约文档(6 篇)
| 文件 | 域 | 场景数 | 优先级 |
|---|---|---|---|
00-master-spec-ai-harness-2026-09-05.md | 总纲 | — | — |
01-spec-software-engineering-2026-09-05.md | A 软件工程域 | 6 | P0 |
02-spec-knowledge-collaboration-2026-09-05.md | B 知识与协同域 | 7 | P0 |
03-spec-risk-compliance-2026-09-05.md | C 风险与合规域 | 5 | P1(紧迫度 P0) |
04-spec-data-science-2026-09-05.md | D 数据与科学域 | 4 | P1 |
05-spec-industry-creative-2026-09-05.md | E 产业与创意域 | 4 | P2(框架版) |
5 大域共 26 个场景,与规划一一对应。
1.2 调研附件(3 份)
research/01-competitive-landscape(竞析)—— 竞品分析与市场格局research/02-user-insights(瑞思)—— 用户洞察与采纳门槛research/03-market-data(数析)—— 市场规模与数据基准
1.3 工具链资产
specs/tooling-schemas-2026-09-05.md—— JSON Schema + MCP 清单 + Eval 集 + deny-rule + CI 集成
1.4 路线图与索引
roadmap-ai-harness-promotion-2026-09-05.md—— 推广路线图(2026 Q4 – 2027 Q3)README.md—— 索引
2. 关键决策(v1.0 锁定)
- 载体三件套:AGENTS.md + SKILL.md + MCP
- 自主度分级:Gartner L1-L4,映射 Google SRE L0-L3
- HITL 按可逆性 × 影响半径插入
- C 域封顶 L3;E 域物理接入封顶 L3
- A 与 B 同批;E 域出框架版
- 引用纪律:回避无原文数字;说明市场规模口径层级
- 北极星指标 WSHR
3. 说明(Notes)
- 5 大域共 26 个场景(与规划一一对应)
- E 域为框架版(数据不足;下版基于真实案例库补齐)
- 北极星指标 WSHR 已建立基线采集机制
- 季度复审机制:下次复审 2026-12-05
4. 后续
两天后(2026-09-07)发布 v1.0.1,新增白皮书 v1.0、行业基准定价评估与利益相关者更新, 规约主体保持不变。详见版本记录 2026 · 0907。
2026 · 0905
Date: 2026-09-05 · Type: version release (initial) · Related version: v1.0.0
The first release of a reusable, industry-segmentable AI Harness spec system — 5 domains, 26 scenarios, each with "usage-spec detail + a copy-ready prompt." Built on the AGENTS.md + SKILL.md + MCP trio, inventing no DSL and binding to no vendor.
1. What Was Released
1.1 Spec Documents (6)
| File | Domain | Scenarios | Priority |
|---|---|---|---|
00-master-spec-ai-harness-2026-09-05.md | Master spec | — | — |
01-spec-software-engineering-2026-09-05.md | A · Software Engineering | 6 | P0 |
02-spec-knowledge-collaboration-2026-09-05.md | B · Knowledge & Collaboration | 7 | P0 |
03-spec-risk-compliance-2026-09-05.md | C · Risk & Compliance | 5 | P1 (urgency P0) |
04-spec-data-science-2026-09-05.md | D · Data & Science | 4 | P1 |
05-spec-industry-creative-2026-09-05.md | E · Industry & Creative | 4 | P2 (framework version) |
26 scenarios across 5 domains, one-to-one with the planned coverage.
1.2 Research Attachments (3)
research/01-competitive-landscape(Competitive) — competitor analysis and market landscaperesearch/02-user-insights(Insights) — user insights and adoption barriersresearch/03-market-data(Market) — market size and data benchmarks
1.3 Tooling Assets
specs/tooling-schemas-2026-09-05.md— JSON Schema + MCP inventory + Eval set + deny-rule + CI integration
1.4 Roadmap & Index
roadmap-ai-harness-promotion-2026-09-05.md— promotion roadmap (2026 Q4 – 2027 Q3)README.md— index
2. Key Decisions (locked in v1.0)
- The carrier trio: AGENTS.md + SKILL.md + MCP
- Autonomy grading: Gartner L1–L4, mapped to Google SRE L0–L3
- HITL inserted by reversibility × impact radius, not by step number
- Domain C capped at L3; Domain E physical access capped at L3
- Domains A and B ship in the same batch; Domain E ships as a framework version
- Citation discipline: avoid figures without source text; always state the market-size metric layer
- North-Star metric: WSHR
3. Notes
- 26 scenarios across 5 domains (one-to-one with the planned coverage)
- Domain E is a framework version (insufficient data; to be completed from a real case library in the next version)
- Baseline collection for the North-Star metric WSHR is in place
- Quarterly review cycle: next review 2026-12-05
4. What's Next
Two days later (2026-09-07), v1.0.1 added the whitepaper v1.0, the benchmark pricing assessment, and the stakeholder update, with the spec body unchanged. See the version record 2026 · 0907.