archeology

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

STARTER_CHARACTER = 🏺

STARTER_CHARACTER = 🏺

Archeology

考古

Reconstruct what happened from surviving evidence, then answer the questions from that record.

Deliverable is

.llmtmp/timeline.md

. Working notes stay under

.llmtmp/

. Write nothing else without being asked.

根据留存证据重建事件经过，然后依据该记录回答问题。

交付物为

.llmtmp/timeline.md

。工作笔记保存在

.llmtmp/

下。未经询问，不得写入任何其他内容。

Input

输入

The argument is one or more questions, split on

. Number them Q1 through Qn and keep that numbering for the whole run. If no argument is given, ask for at least one question and stop until it arrives. Do no research against an unstated question.

参数是一个或多个问题，以

分隔。将其编号为 Q1 至 Qn，并在整个运行过程中保持这一编号。如果未提供参数，请至少询问一个问题，并停止运行，直到收到问题。不得针对未明确提出的问题开展研究。

Sources

来源

These are labels, not instructions. Reach each one with whatever tool the environment provides.

Jira, Confluence, and Slack are the seed. The rest get reached when the evidence points at them. Never ask which sources to use.

Issue trackers and portfolio boards
Wikis, design docs, and decision records
Chat
Mail, calendar, and meeting records
Source repos: commits, pull requests, reviewers, branch policy, pipelines, releases
The code itself
Cloud runtime state
Cost and billing
Logs, traces, monitors, and incident history
Code quality and scan history
Dictionaries and terminology registries
Agent memory and session transcripts
External docs, vendor material, and the web

以下仅为来源标签，并非操作说明。应使用环境提供的任何工具访问每个来源。

Jira、Confluence 和 Slack 是初始来源。证据指向何处，再访问其他来源。绝不询问应使用哪些来源。

问题跟踪器和项目组合看板
Wiki、设计文档和决策记录
聊天
邮件、日历和会议记录
源仓库：commits、pull requests、评审者、分支策略、pipelines、发布
代码本身
云运行时状态
成本和计费
日志、traces、monitors 和事故历史
代码质量和扫描历史
字典和术语注册表
Agent 记忆和会话记录
外部文档、供应商资料和 Web

Procedure

流程

Frame the questions. Restate each question. State the time window the set covers. State the scope boundary, meaning whose story this is and whose it is not.
Seed sweep. Dispatch three subagents in parallel, one each against Jira, Confluence, and Slack. Give each the questions, the subject, the time window, the citation format, and the evidence rules below. These three carry the identifiers everything else needs. Each returns dated, cited findings plus the anchors it confirmed: project keys, epic keys, space keys, page ids, channel ids, repo names, environment names, and every other system the record mentions.
Follow the evidence. Each round names new ground. A ticket names a repo, a page names a dashboard, a message names an incident or a mailbox. Fan out one subagent per newly named source, in parallel, and keep going until a round turns up nothing new. Judge what is worth chasing and chase it. Do not ask which sources to use, and do not stop at the seed. An agent that finds nothing reports the queries it ran and the empty result, which is evidence of absence bounded by that query.
Assemble the milestones. Merge the returned findings into one chronology. Split it into named phases. Tag every entry that bears on a question with
```
[Qn]
```
.
Resolve conflicts. Where a source claim and an artifact disagree, keep both and say which is which. Where two artifacts disagree, keep both and name the discrepancy.
Write the report. Follow the sample below. Same headings, same order, every run.
Record the method. Every access boundary, denial, result cap, and unverified claim. Name the sources reached and the ones the record pointed at that this run never got to.

界定问题。 重述每个问题。说明这组问题涵盖的时间窗口。说明范围边界，即谁的故事属于本次范围，谁的不属于。
初始扫描。 并行分派三个子 Agent，分别针对 Jira、Confluence 和 Slack。向每个 Agent 提供问题、主题、时间窗口、引用格式及以下证据规则。这三个来源提供后续所需的所有标识符。每个 Agent 返回带日期和引用的发现，以及已确认的锚点：project key、epic key、space key、page id、channel id、repo name、environment name，以及记录提到的所有其他系统。
追踪证据。 每一轮都会指出新的调查方向。工单会指向仓库，页面会指向仪表板，消息会指向事故或邮箱。针对每个新发现的来源并行分派一个子 Agent，持续执行，直到某轮没有新发现。判断哪些线索值得追踪并继续追踪。绝不询问应使用哪些来源，也不得停在初始来源。找不到内容的 Agent 应报告执行过的查询和空结果；这是受该查询范围约束的缺失证据。
整理里程碑。 将返回的发现合并为一条时间线。划分为命名阶段。为每个与问题相关的条目添加
```
[Qn]
```
标签。
化解冲突。 当来源主张与工件不一致时，两者都保留，并说明各自是什么。当两个工件不一致时，两者都保留，并指出差异。
撰写报告。 按照以下示例撰写报告。每次运行都使用相同的标题、相同顺序和相同条目形式。
记录方法。 记录每个访问边界、拒绝情况、结果上限和未经验证的主张。注明已访问的来源，以及记录中指向但本次运行未访问的来源。

Sample report

示例报告

Every run produces this shape. Same headings, same order, same entry form.

markdown

undefined

每次运行都生成以下结构。标题、顺序和条目形式保持一致。

markdown

undefined

<Subject>: Evidence Timeline

<主题>：证据时间线

This is a working document. Each milestone carries a slug, a link, and its supporting evidence. Amend as new evidence arrives. Last updated YYYY-MM-DD.

Sources so far: <system>, <system>, and <system>.

这是一份工作文档。每个里程碑都包含 slug、链接和支持证据。有新证据时进行修订。最后更新于 YYYY-MM-DD。

目前使用的来源：<系统>、<系统>和<系统>。

Executive summary

执行摘要

<用连续叙述回答每个问题，使只阅读本节的读者也能理解。按事件顺序展开，并谨慎说明已证实状态。三至六段。>

Summary

摘要

<面向领导层的叙述。每个数据均可追溯到下文某一节。>

<最后一段说明本摘要不主张什么，以及哪些数据存在范围限制。>

Purpose

目的

<N> open questions drive this timeline. Entries are tagged

[Q1]

through

[Qn]

where they bear on one.

Q1. <Question>? <What the record establishes.> <What stays open.>

Q2. <Question>? <What the record establishes.> <What stays open.>

<N> 个开放问题驱动这条时间线。相关条目会标记为

[Q1]

至

[Qn]

。

Q1. <问题>? <记录已经确立的内容。> <仍未确定的内容。>

Q2. <问题>? <记录已经确立的内容。> <仍未确定的内容。>

Acronyms

缩略语

Expansions marked (dict) come from a terminology registry. Expansions marked (meta) come from project or space metadata. Expansions marked (record) are established by artifacts in this timeline. Unverified entries say so.

Term	Expansion	Meaning here
<ABC>	<Expansion> (dict)	<What it refers to in this record>
<XYZ>	<Expansion> (record)	<What it refers to in this record>

标记为 (dict) 的释义来自术语注册表。标记为 (meta) 的释义来自项目或空间元数据。标记为 (record) 的释义由这条时间线中的工件确立。未经验证的条目会明确注明。

术语	释义	此处的含义
<ABC>	<释义> (dict)	<它在这条记录中指代什么>
<XYZ>	<释义> (record)	<它在这条记录中指代什么>

Cast

Person	Role	Established by
<Name>	<Role>	<Artifact and link>

人员	角色	依据
<姓名>	<角色>	<工件和链接>

Milestones

里程碑

Phase 0: <Name>, <start> to <end>

阶段 0：<名称>，<开始> 至 <结束>

YYYY-MM-DD. <Headline>

[Q1]

<ABC-123> | <Page title> <Evidence. Quote the source verbatim where the wording carries the claim.>

YYYY-MM-DD HH:MM TZ. <Headline> <Chat> <Name>: "<verbatim quote>" <What the quote establishes.>

YYYY-MM-DD. <标题>

[Q1]

<ABC-123> | <页面标题> <证据。在措辞承载该主张之处，逐字引用来源。>

YYYY-MM-DD HH:MM TZ. <标题> <聊天> <姓名>: "<逐字引文>" <该引文能够证明的内容。>

Phase 1: <Name>, <start> to <end>, <count> <units>

阶段 1：<名称>，<开始> 至 <结束>，<数量> <单位>

<按日期顺序排列的条目。>

<System> record

<系统>记录

<Metric>	<Value>	Source
<Metric>	<Value>	<Query or link>

<每个系统一个章节：代码仓库事实、问题跟踪器记录、运行时状态、文档记录、消费记录、成本。根据证据需要设置子章节。>

<指标>	<值>	来源
<指标>	<值>	<查询或链接>

Findings

发现

<One claim, carrying the figure that supports it.>
<One claim, carrying the figure that supports it.>

<一项主张，并附上支持该主张的数据。>
<一项主张，并附上支持该主张的数据。>

Open questions

未决问题

<Question.> <What would settle it.>
<Question.> <What would settle it.>

<问题。> <什么信息能够解决它。>
<问题。> <什么信息能够解决它。>

Method

方法

<Query provenance: tool, host, credential, and date for each system.>
<Result caps, and how counts were taken above the expected row count.>
<Denials, recorded rather than worked around.>
<Sources the record named that this run never reached.>
<What stays unverified.>

undefined

<以上每项主张均属于以下三种类型之一：引用来源是作者在相应日期提出的主张，系统记录是工件，实时查询结果是在相应日期观察到的状态。当主张与工件不一致时，两者均予保留。>

<每个系统的查询来源：工具、主机、凭据和日期。>
<结果上限，以及当预期行数较高时如何提高限制后进行计数。>
<拒绝访问的情况，如实记录，不绕过限制。>
<记录中提到但本次运行未访问的来源。>
<仍未验证的内容。>

undefined

Evidence rules

证据规则

Cite the system, the identifier, and the URL on every claim.
A source claim differs from an artifact. An email saying an issue closed is a claim. The issue transition record is the artifact. Say which one a statement rests on.
Record contradictions. Never resolve one silently.
Label every unverified statement as unverified, in place.
Record a denied query as a denial. Do not work around it.
Never print a secret value. Compare and classify without echoing.
All queries are read-only. No write, delete, or restart operation.
Agent memory and transcripts are not independent evidence.
State the result cap on every count, and take counts with a limit above the expected row count.
Bound every count by the credential that produced it.

每个主张都要引用系统、标识符和 URL。
来源主张不同于工件。声称工单已关闭的邮件是一项主张，而工单状态转换记录是工件。说明某项陈述依据的是哪一个。
记录矛盾，绝不擅自消解矛盾。
原地将每个未经验证的陈述标记为未经验证。
被拒绝的查询记录为拒绝，不得绕过。
绝不打印秘密值。只进行比较和分类，不回显。
所有查询均为只读。不得执行写入、删除或重启操作。
Agent 记忆和会话记录不是独立证据。
每个计数都要注明结果上限，并以高于预期行数的限制进行计数。
每个计数都应受产生该计数的凭据权限范围约束。

Framing rules

表述规则

Describe, do not evaluate. "What was requested" is a question. "Whether the request was any good" is a verdict.
Attribute every characterization to its author, verbatim, or delete it.
Do not link sources with convergence language that walks the reader toward a judgment.
Scope every absolute to what was searched, and name the search.
Label list prices, projections, and estimates as what they are.
The clanker-prose plugin's ghostwriting rules apply. The reader is a human colleague.

只描述，不评价。“请求了什么”是问题。“该请求好不好”是裁决。
每项描述性表述都必须逐字归属于其作者，否则删除。
不要使用引导读者作出判断的汇聚性措辞来关联来源。
每个绝对表述都应限定在已搜索范围内，并注明搜索内容。
将标价、预测和估算明确标为各自所属类别。
clanker-prose 插件的代笔规则适用。读者是一位人类同事。

Formatting

格式

ISO dates. Name the timezone when the source gives one.
Wrap prose at 95 columns.
Compact tables: one space inside the pipes, separator dashes sized to the header text.
Do not run
```
md-lint
```
or prettier on the deliverable. Both break the house format.

使用 ISO 日期。来源提供时区时，注明时区。
正文按 95 列换行。
紧凑表格：竖线内保留一个空格，分隔线短横线数量与标题文字长度一致。
不要对交付物运行
```
md-lint
```
或 prettier。二者都会破坏既定格式。