martin1847/evolab / skills/agent-backend-standard
agent-backend-standard
Verified后端架构与工程规范。适用于构建或评审 agent/LLM 后端、任何涉及数据库(DB)的后端,以及建立或评审后端仓库工程门禁。
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 6.5 KB |
| references/agent-foundations.md | markdown | 2.3 KB |
| references/caching-discipline.md | markdown | 3.9 KB |
| references/code-dependency-lifecycle.md | markdown | 9.5 KB |
| references/data-write-discipline.md | markdown | 9.9 KB |
| references/engineering-interface.md | markdown | 10.9 KB |
| references/evals.md | markdown | 2.6 KB |
| references/governance.md | markdown | 2.7 KB |
| references/hitl-resume-safety.md | markdown | 2.7 KB |
| references/observability-cost.md | markdown | 1.2 KB |
| references/prompt-context.md | markdown | 1.6 KB |
| references/resilience.md | markdown | 3.9 KB |
| references/safety-bounded-execution.md | markdown | 4.5 KB |
| references/secrets-discipline.md | markdown | 2.5 KB |
| references/selfcheck-gates.md | markdown | 4.4 KB |
| references/state-durability.md | markdown | 2.6 KB |
| references/tool-design.md | markdown | 2.8 KB |
references/selfcheck-gates.md
View on GitHub# 附录 F — 自检清单与评审蒸馏门禁(selfcheck gates)<!-- trunk:selfcheck-gates.md -->
> **电在回路(shock-in-the-loop):soft prompts steer, hard gates hold the line。**
> canonical:`cto-orchestration/references/shock-in-the-loop.md`(下沉判据/分层/DENY 三件套/
> override 有形;概念为本工程体系首创)。
> 写作纪律**「能电不文」**:能下沉的绝不写散文;散文只留方法论与门判不了的判断项(可稍详不啰嗦)。
>
> 定位:门禁**从哪来、如何演化**。门禁接口长什么样归附录 C(engineering-interface),
> 本章管的是另一条供给线:把真实评审 findings 蒸馏成仓库自己的护栏。
> 实证来源:某双服务后端仓活体运行(五条负探针实测;曾实抓「契约语义收窄未升版」
> 「聚合数组漏登静默失守」两类真事故)+ 某下游仓 SPEC 镜像漂移 / docs-site 错分支两起真事故。
## 回路(机制本体)
```
真实评审 finding / 真事故
→ 蒸馏成 docs/PR_SELF_CHECK.md 条目(作者提 PR 前过一遍,人和 agent 同一份)
→ 其中可机检的下沉 diff-scoped pre-push hook(毫秒级拦截)
→ 每关带负探针,登记进门禁台账(AGENTS.md 验证配方节)
→ 下一个真 finding 继续喂清单
```
价值主张:把评审往返从多轮压到 1 轮;清单喂人也喂 agent(AGENTS.md 指过去,
agent 提 PR 前自查)。
## 铁律
- **hook 里每一关必须能指到一次真实事故或评审 finding;禁止凭理论发明关卡。**
这是本机制与一般 lint 清单的本质区别,也是防清单腐烂成仪式的机制。
新仓落地时清单可以为空——先攒 findings 再蒸馏,不空写通用建议(同 hub 的 TBD 纪律)。
## 下沉判据(清单条目 → hook 关卡)
全部满足才下沉,否则留在清单里靠人/agent 自查:
1. **可机械判定**:纯 git + grep/sed 能出确定性结论,无需理解语义。
2. **毫秒级**:不跑构建、不跑测试、不碰网络(构建级检查归附录 C 的 check/test 或 CI)。
3. **diff-scoped**:只看本次 push 的 range,不做全仓扫描(range walk 骨架 +
`git config hooks.baseline` 参数化基线分支,见 bootstrap skill 的 `pre-push.template`)。
## hook 是提醒、CI 是门
pre-push 可被 `--no-verify` 绕过,本质是**快速反馈层**(write-time 预防,软层)。
脚本必须设计成可带 range 参数独立调用(如 `pre-push --range origin/main..HEAD`),
CI 复跑同一脚本才构成真门(硬层归 IaC / required CI,与附录 C「本地 hook 不冒充
required CI」同款双层)。
## 门的适用性纪律(绿 ≠ 门跑过)
CI 的 workflow 定义取自 **PR head commit**:head 早于某道门落地的 PR,其 CI 里该门
**根本不存在**——「全绿」对该门零证据力,`rerun` 也无效(复用旧定义)。下游实证
(2026-07-29):门落地当天 5 个 open PR 全绿,ancestry 核查后 2 个一道新门都没跑过。
机械判据(可下沉进合并政策):`git merge-base --is-ancestor <门合入 commit> <PR head>`,
非 ancestor ⇒ rebase 后门才第一次真跑。同族条目:「绿因为零覆盖」讲规则空转,
本条讲**运行本身没发生**——门落地日必须清点存量 open PR 的盲区。
## 负探针台账
每关落地时同车交付负探针配方(怎么让它红、红完怎么恢复),登记进该仓 AGENTS.md
的验证配方节(无此节则建)。**没有实测过 exit 1 的关卡不算存在**——与「验证诚实性」
同源:门禁自身也要正向证据。
## 与附录 C 的分工
| | 附录 C engineering-gate | 附录 F selfcheck gates |
| --- | --- | --- |
| 供给线 | 语言/工具链静态质量(lint/type/test) | 评审 findings / 事故蒸馏 |
| 挂载点 | pre-commit + CI | pre-push + CI |
| 内容来源 | 语言规范(附录 C §1–§6 canonical) | 本仓真实教训,仓仓不同 |
| 速度契约 | check/test 级 | 纯 git+grep 毫秒级 |
## 落地路径
- 新仓:`repo-governance-bootstrap` 建骨架(`pre-push.template` + `PR_SELF_CHECK`
skeleton + AGENTS.md 指引 slot),清单从空开始。
- 存量仓:从最近一季真 findings 里蒸馏首批条目;可机检的按下沉判据挑 1–2 关起步。
- 下游生态可留面向自己的自包含快照(公开仓不能引私有 skill);方法论演化以本章为
canonical,快照机会性同步。
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |