evidence-docs
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseEvidence-backed documentation
基于证据的文档
A claim is documentation only when it carries the means to check it. This skill is
the standard and the map: ten canons, and where each is defined, enforced and seeded.
It is a navigator, not a second copy. Every law below has exactly one home — that is
canon 3, and a navigator that restated the doctrine would break the rule it is indexing.
The full statement of each canon, its rationale and its enforcement live in
→ The canons.
documentation.md只有当内容附带可验证的依据时,才能称之为文档。 本技能是标准与指引:包含十条准则,以及每条准则的定义、执行方式和来源。
它是一个导航工具,而非重复副本。以下每条准则都有唯一的来源——这正是准则3的要求,若导航工具重述规范,便会违背其索引的规则。每条准则的完整说明、原理及执行细节请查看 → 准则部分。
documentation.mdThe ten canons
十条准则
- A claim carries its address — , a command with its output, a test name; a lesson names its commit.
file:line - Numbers are computed, never restated.
- Every fact has exactly one home — others link, never restate.
- A reference resolves from where the document is read — not from where it lives.
- Green nobody watched turn red is not evidence.
- A check proves its scope and nothing beyond it.
- Silence is not a pass — ask what a mechanism prints when it did not look.
- An estimate is never announced as a measurement — a rule states its evidence condition.
- What was not checked is printed beside what was.
- The document ships in the change that made it true — and a correction is appended, never written over.
They are epistemic: what makes a claim documentation. The operational layer — what to
do at a given trigger, with a check and an exit criterion — is
. When the two seem to say the same
thing, the canon is the why and the rule is the how.
learned.md- 内容附带定位信息 —— 、带输出的命令、测试名称;经验教训需标注对应的提交记录。
file:line - 数据需计算得出,不得重复转述。
- 每个事实仅有唯一来源 —— 其他内容仅可链接,不得重复转述。
- 引用需从文档阅读位置解析 —— 而非文档存储位置。
- 无人关注的成功结果不算证据。
- 验证仅能证明其覆盖范围,不涉及其他内容。
- 无反馈不等于通过 —— 需明确机制未检查时的输出内容。
- 估算不得当作实测结果发布 —— 规则需明确其证据条件。
- 未检查内容需与已检查内容一并列出。
- 文档随使其生效的变更一同发布 —— 修正内容需追加,不得覆盖原文。
这些是认知层面的要求:界定了何为有依据的文档。操作层面——即特定触发场景下的行动、验证方式及退出标准——请查看。当两者内容看似重叠时,准则是原因,规则是执行方法。
learned.mdWhere next
下一步指引
| You are about to… | Read | Because |
|---|---|---|
| set a project's documentation up from nothing | | four questions answered before the first line of work |
| record a decision so it survives its author | Registers and ids + | append-only ids, edge markers, one decision home |
| change something and not orphan the docs | The Doc Loop + The propagation matrix | which documents a change owes, starting with the meta-row |
| decide where a fact belongs | Single source of truth | two homes disagree the day one of them is updated |
| build a check that cannot lie | | three axes, the enforcement ladder, progressive arming, probing |
| trust a mechanism that reports success | | the failure that removes the reason to look |
| wire a check into the agent's own tooling | | the hook contract, and why a crashed guard allows the action |
| audit documentation a project already has | | seven passes, cheapest first, output is a fix plan |
| carry a lesson to the next run | | stamp first (the cold trigger reads it), then prune to a cap of ten; every lesson names its commit |
| seed a gate into a host project | | it seeds green: dormant where there is no input yet |
| claim that an agent behaves | | the address is a trace id and the assertion that ran (canon 1); a judge nobody watched disagree is a green nobody watched turn red (canon 5) |
| take a whole change through to acceptance | | this skill is the standard; that one is how a change reaches the repository |
| 你即将要… | 阅读内容 | 原因 |
|---|---|---|
| 从零开始搭建项目文档体系 | | 开始工作前需明确四个核心问题 |
| 记录决策以确保其不受作者离职影响 | 登记册与ID + | 仅追加式ID、边界标记、决策唯一来源 |
| 变更内容且不导致文档脱节 | 文档循环 + 传播矩阵 | 变更需同步的文档范围,从元数据行开始 |
| 确定事实的归属位置 | 单一事实来源 | 若存在两个来源,其中一个更新时必然出现分歧 |
| 构建可靠的验证机制 | | 三个维度、执行层级、渐进式启用、探查机制 |
| 信任报告成功的机制 | | 此类失败会消除进一步检查的必要性 |
| 将验证机制接入Agent工具链 | | 钩子合约,以及为何崩溃的防护会允许操作执行 |
| 审计项目现有文档 | | 七个检查步骤,从成本最低的开始,输出修复方案 |
| 将经验教训带入下一轮工作 | | 先标记(冷触发时读取),再精简至最多十条;每条经验教训需标注对应的提交记录 |
| 在宿主项目中植入文档准入机制 | | 初始状态为绿色:无输入时处于休眠状态 |
| 声称Agent行为合规 | | 定位信息为追踪ID及执行的断言(准则1);无人关注的判定分歧属于“无人关注的成功结果变失败”(准则5) |
| 完成完整变更的验收流程 | | 本技能是标准,而该技能是变更进入代码库的流程指引 |
When this applies
适用场景
The boundary is "it will be read as true." A decision record, a README, an acceptance
report, a runbook, a changelog for users, an audit finding, a claim that something was
verified.
Not through this skill: a draft, thinking out loud, an answer in chat, a commit
message, a code comment. Demanding a for "let me check that" is the fastest
way to teach an agent to route around the rule where it actually protects something.
file:lineRefusal phrase — "без доков" / "on my word". It works on a task that would otherwise
pass through here: do it directly and say out loud that the claim is unbacked, rather
than presenting an estimate as a measurement (canon 8).
边界是“内容会被当作事实”。包括决策记录、README、验收报告、运行手册、面向用户的变更日志、审计结果、声称已验证的内容。
不适用场景:草稿、即兴想法、聊天回复、提交信息、代码注释。若对“我确认一下”这类内容要求定位,会导致Agent在规则实际需要保护的场景下绕开规则。
file:line拒绝用语 —— "без доков" / "on my word"。此用语可用于原本会进入本流程的任务:直接执行操作,并明确说明内容无依据,而非将估算当作实测结果发布(准则8)。
The one test
唯一测试标准
Before a document ships, read it for the sentence that would embarrass you if someone
asked "how do you know?" — and either give that sentence its address, or delete it.
Everything above is that question, made mechanical.
文档发布前,通读并找出若有人问*“你怎么知道的?”*会让你尴尬的句子——要么为该句子添加定位依据,要么删除它。以上所有规则都是将这个问题转化为可执行的机械流程。