eiirp

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

EIIRP — Everything In Its Right Place

EIIRP — 各归其位（Everything In Its Right Place）

"Everything in its right place" — Radiohead, Kid A

"各归其位" — Radiohead，《Kid A》

Contract

约定

After any significant work, EIIRP organizes ALL outputs across two domains:

Knowledge domain (brain):

Every piece of knowledge lands in the correct brain location.
All sources are cited and linked.
The active schema pack is updated if a new content type emerged.
Entity pages created/updated with cross-links.

Capability domain (skills): 5. Every reusable pattern becomes a composable skill. 6. Existing skills are audited for DRY violations. 7. Skill graph is MECE — no gaps, no overlaps, no ambiguous routing.

The meta-guarantee: Nothing produced during significant work lives only in chat. Knowledge → brain. Patterns → skills. Everything in its right place.

在完成任何重要工作后，EIIRP会整理两个领域的所有产出物：

知识域（brain）：

每一项知识都被放置到brain的正确位置。
所有来源都被引用并添加链接。
如果出现新的内容类型，更新当前使用的schema pack。
创建/更新实体页面并添加交叉链接。

能力域（skills）： 5. 每一个可复用模式都成为可组合的skill。 6. 审核现有skills是否违反DRY原则。 7. 技能图谱符合MECE原则——无遗漏、无重叠、无模糊路由。

元保障： 重要工作中产生的内容不会仅存于聊天记录中。知识存入brain，模式转化为skills。所有内容各归其位。

When to Use

使用场景

After completing a deep research thread.
After building something new (code, pipeline, workflow).
After a multi-source analysis that produced significant findings.
When the user says "EIIRP", "organize this", "DRY this up", "make this re-doable".
When a work session produced both knowledge AND new capabilities.
When you notice skill overlap, duplication, or gaps.

完成深度研究线程后。
构建新事物（代码、流水线、工作流）后。
完成多源分析并产生重要结论后。
用户说出"EIIRP"、"整理这个"、"优化重复内容"、"让这个可复用"时。
工作会话同时产生知识和新能力时。
发现skill重叠、重复或存在缺口时。

Phase 1: INVENTORY — What did we produce?

阶段1：清点——我们产出了什么？

Scan the current session/thread and identify ALL outputs across both domains.

扫描当前会话/线程，识别两个领域的所有产出物。

Knowledge outputs

知识产出物

□ Primary findings (the synthesis)
□ Source documents (URLs, PDFs, articles, tweets)
□ Entity mentions (people, companies, organizations, places)
□ Concepts/frameworks (reusable mental models)
□ Data artifacts (structured data, timelines, statistics)

□ 核心结论（综合成果）
□ 来源文档（URL、PDF、文章、推文）
□ 实体提及（人物、公司、组织、地点）
□ 概念/框架（可复用思维模型）
□ 数据工件（结构化数据、时间线、统计数据）

Capability outputs

能力产出物

□ New skills created or modified
□ Scripts/code written (should they be in lib/ or scripts/?)
□ Methodology used (search patterns, source chains, verification steps)
□ Workflows that could be automated (cron, pipeline, webhook)
□ Patterns that will recur (→ candidate for skillification)

Produce a manifest:

markdown

undefined

□ 创建或修改的新skills
□ 编写的脚本/代码（应放入lib/还是scripts/?）
□ 使用的方法论（搜索模式、来源链、验证步骤）
□ 可自动化的工作流（cron、流水线、webhook）
□ 会重复出现的模式（→ 可转化为skill的候选）

生成清单：

markdown

undefined

EIIRP Manifest

EIIRP清单

Topic: [topic]
Date: [date]
Knowledge outputs: [count] (sources, entities, concepts)
Capability outputs: [count] (skills, scripts, patterns)
Reusable methodology: [yes/no — describe if yes]

undefined

主题：[主题]
日期：[日期]
知识产出物：[数量]（来源、实体、概念）
能力产出物：[数量]（skills、脚本、模式）
可复用方法论：[是/否 — 若是请描述]

undefined

Phase 2: TAXONOMY — Where does each piece go?

阶段2：分类体系——每一项内容归属于哪里？

Read the active schema pack first (the single source of truth for filing decisions in v0.39+):

bash

gbrain schema show --json

The pack's

page_types[]

lists every directory the brain accepts plus the primitive each maps to. Walk it for each output and pick the directory whose

path_prefixes

matches the content's primary subject.

brain-taxonomist

is installed, INVOKE IT for ambiguous cases. It runs the same decision protocol against the active pack and gives you a single recommended filing path with reasoning.

Output: a filing plan table:

| Content | Brain path | Action |
|---------|-----------|--------|
| Primary research | reference/.../page.md | CREATE |
| Person X | people/x-slug.md | CREATE |
| Person Y | people/y-slug.md | UPDATE (already exists) |
| ... | ... | ... |

先阅读当前使用的schema pack（v0.39+版本中归档决策的唯一事实来源）：

bash

gbrain schema show --json

该pack的

page_types[]

列出了brain接受的所有目录及其对应的基础类型。针对每个产出物梳理分类体系，选择

path_prefixes

与内容主题匹配的目录。

如果已安装

brain-taxonomist

，在遇到模糊情况时调用它。它会根据当前pack执行相同的决策流程，并给出带推理过程的唯一推荐归档路径。

输出：归档计划表：

| 内容 | Brain路径 | 操作 |
|---------|-----------|--------|
| 核心研究内容 | reference/.../page.md | 创建 |
| 人物X | people/x-slug.md | 创建 |
| 人物Y | people/y-slug.md | 更新（已存在） |
| ... | ... | ... |

Phase 3: SCHEMA CHECK — Does the active pack cover this content?

阶段3：Schema检查——当前pack是否覆盖该内容？

This is where EIIRP closes the schema-derivation loop. If the work produced content that doesn't fit any existing

page_types

, propose adding a new type via the v0.39 cathedral:

bash

undefined

这是EIIRP闭合schema推导循环的环节。如果工作产出的内容不符合任何现有

page_types

，通过v0.39 cathedral提议添加新类型：

bash

undefined

What's emerging in the brain that the active pack doesn't cover?

brain中出现了哪些当前pack未覆盖的内容？

gbrain schema detect --json

LLM-refined suggestions (heuristic when no API key set).

LLM优化的建议（未设置API密钥时的启发式方案）。

gbrain schema suggest --json

Review what's pending; promote or ignore each candidate.

查看待处理候选；批准或忽略每个候选。

gbrain schema review-candidates --json gbrain schema review-candidates --apply <prefix-or-type-name>


**Confidence floor (codex finding #9):** when `gbrain schema suggest`
returns confidence < 0.6 on a proposed type, DO NOT auto-apply. Surface
the suggestion to the user and let them choose. The schema-cathedral
ships the primitives; EIIRP enforces the human-in-the-loop gate.

If schema needs change:
- Propose the addition to the user before running `review-candidates --apply`.
- Document the change in the commit message of the next sync.
- The schema-pack engine writes the delta to
  `~/.gbrain/schema-pack-deltas/` — review and merge into the active
  pack via `gbrain schema edit` (or hand-edit the YAML).

gbrain schema review-candidates --json gbrain schema review-candidates --apply <prefix-or-type-name>


**置信度下限（法典发现#9）：** 当`gbrain schema suggest`返回的提议类型置信度<0.6时，请勿自动应用。将该建议提交给用户，由用户选择。schema-cathedral提供基础类型；EIIRP强制执行人工审核环节。

若需要修改schema：
- 在执行`review-candidates --apply`前，向用户提议添加新类型。
- 在下次同步的提交信息中记录该变更。
- schema-pack引擎会将差异写入`~/.gbrain/schema-pack-deltas/` — 通过`gbrain schema edit`（或手动编辑YAML）审核并合并到当前pack中。

Phase 4: FILE — Create enriched brain pages

阶段4：归档——创建经过丰富的brain页面

For each item in the filing plan:

针对归档计划表中的每一项：

4a. Primary research page

4a. 核心研究页面

Use the brain page template. MUST include:

Proper frontmatter (
```
type
```
,
```
title
```
,
```
date
```
,
```
tags
```
, sources)
State section — current status/key findings
Sources section — every source with URL, author, date, language
Timeline section — chronological development
Entity links — backlinks to all related brain pages
See Also — related concepts, reference pages

使用brain页面模板。必须包含：

正确的前置元数据（
```
type
```
、
```
title
```
、
```
date
```
、
```
tags
```
、来源）
状态部分——当前状态/核心结论
来源部分——每个来源的URL、作者、日期、语言
时间线部分——按时间顺序的发展历程
实体链接——指向所有相关brain页面的反向链接
另请参阅——相关概念、参考页面

4b. Entity pages (people, companies)

4b. 实体页面（人物、公司）

For each entity mentioned:

Check if a brain page exists (

gbrain search "<name>"

gbrain get_page people/<slug>

If exists: update State, append Timeline entry citing this research.
If not: create with enrichment.

针对每个提及的实体：

检查是否存在brain页面（

gbrain search "<名称>"

或

gbrain get_page people/<slug>

）。

若存在：更新状态，添加引用本次研究的时间线条目。
若不存在：创建并丰富页面内容。

4c. Commit and verify

4c. 提交并验证

After ALL pages are written, run

gbrain sync

(or commit + push in the brain repo). Verify every link resolves.

完成所有页面编写后，执行

gbrain sync

（或在brain仓库中提交+推送）。验证所有链接均可解析。

Phase 5: SKILL GRAPH AUDIT — DRY + MECE on capabilities

阶段5：技能图谱审核——能力域的DRY + MECE检查

This phase operates on the SKILL graph, not just the research.

本阶段针对SKILL图谱而非仅研究内容进行操作。

5a. New pattern identification

5a. 识别新模式

Ask: did this work reveal REPEATABLE patterns that will recur?

Indicators of a reusable pattern:

You used a specific sequence of searches across multiple sources.
You followed a specific verification/cross-referencing methodology.
You wrote code that could be parameterized for different inputs.
The output format is generalizable.
The user is likely to ask for similar work on a different topic.

For each identified pattern:

Identify the composable pieces (DRY, MECE):
- Shared logic →
```
lib/
```
  (not copy-pasted into skills)
- Search methodology → skill or lib function
- Output template → brain template or skill phase
- Filing logic → already covered by brain-taxonomist + active pack
DRY check via the v0.19 resolver:
bash
```
gbrain check-resolvable
```
Look for overlapping triggers or unreachable skills.

思考：本次工作是否揭示了会重复出现的可复用模式？

可复用模式的指标：

你在多个来源中使用了特定的搜索序列。
你遵循了特定的验证/交叉引用方法论。
你编写的代码可针对不同输入进行参数化。
输出格式可通用化。
用户可能会针对不同主题要求类似的工作。

针对每个识别出的模式：

识别可组合的组件（符合DRY、MECE原则）：
- 共享逻辑 →
```
lib/
```
  （不要复制粘贴到skills中）
- 搜索方法论 → skill或lib函数
- 输出模板 → brain模板或skill阶段
- 归档逻辑 → 已由brain-taxonomist + 当前pack覆盖
通过v0.19解析器进行DRY检查：
bash
```
gbrain check-resolvable
```
查找重叠触发器或无法访问的skills。

5b. Existing skill audit

5b. 现有skill审核

For ALL skills used or touched during this work, check:

Were any skills BYPASSED? (did you do something manually that a skill should handle?)
Are there skills that OVERLAP with what you just did? (merge candidates)
Is shared code copy-pasted between skills? (extract to
```
lib/
```
)

The MECE question: If someone asked for this exact work again tomorrow on a different topic, which skills would they invoke? Is the path clear and unambiguous? If not, fix the routing.

针对本次工作中使用或涉及的所有skills，检查：

是否有skills被绕过？（你手动完成了某个skill应处理的工作？）
是否有skills与你刚完成的工作重叠？（合并候选）
skills之间是否存在复制粘贴的共享代码？（提取到
```
lib/
```
）

MECE问题： 如果明天有人针对不同主题要求完全相同的工作，他们会调用哪些skills？路径是否清晰明确？若否，修复路由。

5c. Present the plan

5c. 提交计划

undefined

undefined

Skill Graph Changes

技能图谱变更

New skills to create

待创建的新skills

[skill-name] — [what it does]
- DRY check: [clean / overlaps with X]
- Recommendation: [create / merge into X]

[skill名称] — [功能描述]
- DRY检查：[无重复 / 与X重叠]
- 建议：[创建 / 合并到X]

Existing skills to update

待更新的现有skills

[skill-name] — [what changed, why]

[skill名称] — [变更内容及原因]

Code to extract to lib/

待提取到lib/的代码

lib/[name].ts — [what it does, which skills use it]

lib/[名称].ts — [功能描述，涉及哪些skills]

Skills to merge or deprecate

待合并或弃用的skills

[skill-A] + [skill-B] → [merged-skill] — [why]


On approval: invoke `/skillify` for each new/modified skill.

[skill-A] + [skill-B] → [合并后的skill] — [原因]


获得批准后：针对每个新/修改的skill调用`/skillify`。

Phase 6: CHECK_RESOLVABLE — Verify everything routes

阶段6：可解析性检查——验证所有路由

After all filing and skillification:

bash

gbrain check-resolvable                         # routing-table reachability
gbrain doctor --json                            # health surface
gbrain search "<topic keywords>"                # brain pages findable
gbrain orphans                                  # any pages without inbound links?

Confirm:

All brain pages have proper frontmatter against active schema pack
All entity pages are cross-linked
Any new skills have routing entries in
```
skills/RESOLVER.md
```
No DRY violations (no duplicated logic across skills)
No MECE violations (no ambiguous routing between skills)
Active schema pack updated if new content types emerged

gbrain doctor

reports

schema_pack_consistency: ok

完成所有归档和skill转化后：

bash

gbrain check-resolvable                         # 路由表可达性
gbrain doctor --json                            # 健康状态检查
gbrain search "<主题关键词>"                # brain页面可搜索性
gbrain orphans                                  # 是否存在无入链的页面？

确认：

所有brain页面均包含符合当前schema pack的前置元数据
所有实体页面均已添加交叉链接
所有新skills均在
```
skills/RESOLVER.md
```
中有路由条目
无DRY违规（skills间无重复逻辑）
无MECE违规（skills间无模糊路由）
若出现新内容类型，已更新当前schema pack

gbrain doctor

报告

schema_pack_consistency: ok

Phase 7: REPORT — Summary

阶段7：报告——总结

markdown

undefined

markdown

undefined

EIIRP Complete: [Topic]

EIIRP完成：[主题]

Brain pages created/updated

创建/更新的brain页面

[path] — [description]
...

[路径] — [描述]
...

Entity pages

实体页面

[path] — [created/updated]
...

[路径] — [创建/更新]
...

Schema changes

Schema变更

[none / description of changes + which pack delta file]

[无 / 变更描述 + 对应的pack差异文件]

Skills identified

识别出的skills

[skill-name] — [status: created / merged / deferred]
...

[skill名称] — [状态：已创建 / 已合并 / 已推迟]
...

Resolver status

解析器状态

DRY check: [clean]
MECE audit: [clean]
Active pack: [name] v[version]
schema_pack_consistency: [ok / warn — pct untyped]

undefined

DRY检查：[无违规]
MECE审核：[无违规]
当前pack：[名称] v[版本]
schema_pack_consistency: [正常 / 警告 — 未分类内容占比]

undefined

Output Format

输出格式

EIIRP produces a single Phase 7 report block. Plain markdown:

markdown

undefined

EIIRP生成单个阶段7报告块。纯markdown格式：

markdown

undefined

EIIRP Complete: [topic]

EIIRP完成：[主题]

Brain pages created/updated

创建/更新的brain页面

[path] — [description]

[路径] — [描述]

Entity pages

实体页面

[path] — [created|updated]

[路径] — [创建|更新]

Schema changes

Schema变更

[none | description of changes + which pack delta file]

[无 | 变更描述 + 对应的pack差异文件]

Skills identified

识别出的skills

[skill-name] — [status: created|merged|deferred]

[skill名称] — [状态：已创建|已合并|已推迟]

Resolver status

解析器状态

DRY check: [clean|N violations]
MECE audit: [clean|N overlaps]
Active pack: [name] v[version]
schema_pack_consistency: [ok|warn — N% untyped]


Always machine-readable: stable section headers + bullet-per-item. The
report doubles as a sync checkpoint for downstream skills (skillpack-check
reads it; doctor cross-references the pack version).

DRY检查：[无违规|N项违规]
MECE审核：[无违规|N项重叠]
当前pack：[名称] v[版本]
schema_pack_consistency: [正常|警告 — N%未分类]


始终保持机器可读：固定的章节标题 + 每项内容对应一个项目符号。该报告同时作为下游skills的同步检查点（skillpack-check会读取它；doctor会交叉引用pack版本）。

Anti-Patterns

反模式

Hardcoding directory tables in EIIRP's logic. Every filing decision reads
```
gbrain schema show --json
```
. Users on
```
gbrain-recommended
```
AND custom packs MUST get the right behavior automatically. Pinned by D9 from /plan-eng-review.
Auto-applying low-confidence schema suggestions. Confidence < 0.6 from
```
gbrain schema suggest
```
is "manual review required" per codex finding #9. EIIRP surfaces it; the user accepts.
Skipping Phase 5 SKILL GRAPH AUDIT because "this was a one-off." If the work took >10 minutes, the methodology is probably reusable. Audit anyway; defer the skillify decision to the user.
Filing synthesis output by topic alone. Synthesis pages tied to a single source + reader are sui generis; they file under
```
media/<format>/<slug>-personalized.md
```
. See _brain-filing-rules.md "Sanctioned exception" section.
Treating non-English sources as secondary citations. Multilingual sources are first-class.

在EIIRP逻辑中硬编码目录表。 所有归档决策均读取
```
gbrain schema show --json
```
。使用
```
gbrain-recommended
```
和自定义pack的用户必须自动获得正确行为。由/plan-eng-review中的D9约束。
自动应用低置信度的schema建议。 根据法典发现#9，
```
gbrain schema suggest
```
返回置信度<0.6时需"人工审核"。EIIRP仅提交建议，由用户确认。
因"这是一次性工作"而跳过阶段5技能图谱审核。 如果工作耗时超过10分钟，其方法论很可能可复用。无论如何都要审核，将skill转化决策推迟给用户。
仅按主题归档综合产出物。 与单一来源+读者绑定的综合页面属于特殊情况；应归档到
```
media/<格式>/<slug>-personalized.md
```
。请参阅_brain-filing-rules.md的"认可例外"章节。
将非英文来源视为次要引用。 多语言来源是一等公民。

Hard Rules

硬性规则

Knowledge domain

知识域

Never leave research only in chat. If it took >10 minutes to produce, it gets a brain page.
Every source gets a citation. No "according to reports" without a URL.
Entity pages get updated, not just created. If a brain page exists, UPDATE it.
Schema changes require confirmation. The active pack is load-bearing.
Multilingual sources are first-class. Never treat non-English sources as secondary.

绝不让研究内容仅存于聊天记录中。 如果产出耗时超过10分钟，必须创建brain页面。
每个来源都必须被引用。 禁止无URL的"据报道"表述。
实体页面需更新而非仅创建。 如果brain页面已存在，必须更新它。
Schema变更需确认。 当前pack是核心支撑。
多语言来源是一等公民。 绝不能将非英文来源视为次要。

Capability domain

能力域

DRY is sacred. If the same logic appears in two skills, extract it to
```
lib/
```
.
MECE is sacred. Every trigger phrase routes to exactly one skill.
Composability over monoliths. Small skills that compose > one giant skill that does everything.
Skillify only what recurs. One-off work doesn't need a skill. Patterns that repeat 2+ times do.

DRY原则不可侵犯。 如果相同逻辑出现在两个skills中，需提取到
```
lib/
```
。
MECE原则不可侵犯。 每个触发短语仅路由到一个skill。
组合性优于单体化。 可组合的小型skills > 包办一切的巨型skill。
仅将重复出现的内容转化为skill。 一次性工作无需skill；重复2次及以上的模式需要。

元规则

EIIRP is idempotent. Running it twice on the same work should produce no changes the second time.
EIIRP consumes the active schema pack as data. Never hard-code directory tables in EIIRP's logic — read from
```
gbrain schema show --json
```
so users who picked
```
gbrain-recommended
```
OR custom packs get the right behavior automatically.

EIIRP具有幂等性。 对同一工作运行两次EIIRP，第二次不应产生任何变更。
EIIRP将当前schema pack作为数据使用。 绝不在EIIRP逻辑中硬编码目录表——读取
```
gbrain schema show --json
```
，确保使用
```
gbrain-recommended
```
或自定义pack的用户都能自动获得正确行为。

Changelog

更新日志

v1.0.0 — gbrain v0.39.0.0

Initial port from upstream OpenClaw. Genericized — no references to private fork names per CLAUDE.md privacy rules.
Phase 3 SCHEMA CHECK rewritten to consume the v0.39 cathedral CLI (
```
detect | suggest | review-candidates
```
) instead of a private
```
brain/schema.md
```
.
Phase 5 SKILL GRAPH AUDIT calls
```
gbrain check-resolvable
```
instead of upstream
```
scripts/skill-dry-check.mjs
```
.
Phase 6 verification uses
```
gbrain doctor
```
's schema_pack_consistency check (T7) for the persistent surface.

从上游OpenClaw首次移植。通用化处理——根据CLAUDE.md隐私规则，移除所有私有分支名称引用。
阶段3 Schema检查重写为使用v0.39 cathedral CLI（
```
detect | suggest | review-candidates
```
），而非私有
```
brain/schema.md
```
。

阶段5技能图谱审核调用

gbrain check-resolvable

，而非上游

scripts/skill-dry-check.mjs

。

阶段6验证使用
```
gbrain doctor
```
的schema_pack_consistency检查（T7）作为持久化状态检查。