ljg-rank
Given a domain, identify the few independent forces genuinely underpinning it. Prune dozens of phenomena down to an irreducible set of generators—only counted valid if all phenomena can be regenerated back from them. Use when user says '降秩', '找秩', '秩是什么', '这个领域靠什么撑着', '背后是什么', or wants to decompose any domain to its irreducible generators.
1How It Works & Mechanism
Rank = genuinely independent generators in the domain; evaluate using the two criteria from 'The Beginning of Infinity' (reach / hard to vary).
**解释力**(reach)——不光能推出清单里的现象,还能推出清单外的,而且现实里真验得上。
Look upward before rank reduction: unearth the domain's fundamental assumptions (the ceiling)—the axioms that cannot be questioned and hold true only through belief.
**这里什么是不许问的?**(一问就被当外行,或者被当冒犯)
Distinguish fundamental assumptions from root rank: assumptions form the ceiling (stood up by belief), while root rank forms the floor (revealed through digging).
**root rank 是地板**——一层一层往下追,追到追不动,挖而见。
The seven-step penetration process, centered on Step 3 recursive descent: establish solid footing layer by layer without skipping steps.
**先只找第一层生成器——现象底下紧挨着的、看得见够得着的那层。找齐了,把这一层当新地面,人站上去。**
Step 7 dual testing: out-of-catalog prediction (reach) + variance testing (hard to vary); it is a true root only if changing one part collapses the whole construct.
**变更测试**(hard to vary):随便挑一根,改一个细节——换个条件、换个方向、换个强度。预测还对吗?还对,这根是松的伪根,可以换;改一处就垮,这根才是紧的真根,动不得。
Review the five internal criteria (generative / minimal / independent / reach / hard to vary) post-analysis without writing them into the article.
工序走完之后,这五条是事后再过一遍的闸,**同样不是写出来给人看的**。
Two-layer judgment: first determine the root rank topology (pillars / nested / cyclic / chain), then assess whether an operational coordinate system can be overlaid.
先别急着写——分两层看它们长成什么样。
Anti-collapse gate: rule out all nine topologies one by one before settling on an architecture, explicitly flagging AI lazy anchors (vertical 3-layer profiles and 2x2 grids).
**AI 的另一个懒惰锚点是 2x2 矩阵**。
Native Chinese anti-collapse gate: map anglicized phrasing against natural Chinese equivalents; if the prompt question yields 'would not say this naturally', scrap the paragraph and rethink.
下笔前,默念三遍:**「一个没读过英文的中国人,会这么说话吗?」**
Include an ASCII structure diagram at the end, shaped strictly by the skeleton; enforce pure ASCII with a hard constraint banning all Unicode symbols.
**硬约束：只用纯 ASCII 字符**。禁用任何 Unicode 符号（包括箭头 → ← ↑ ↓、方框 ┌─┐└┘├┤│、圆点 • ◆ ●、粗体 ▶ ◀ 等）。
2Core Capabilities
3External Dependencies
| Kind | Dependency / Endpoint |
|---|---|
| cli | date 取时间戳（输出节命名用） |
4Recommended Governance & Evidence
- 本地工作目录: File I/O and artifact output: Written by the agent to ~/Documents/notes/{timestamp}--{domain}的秩__rank.org per output contract; zero disk writes by the skill directory itself.
5Second-Pass Verification
- [ok] 实现原理＝纯 prompt 编排 + 一套自检判据，无自带可执行代码 — 目录仅 SKILL.md 与 references/template.org；token 扫描 curl/fetch/child_process/process.env 零命中，『执行外部程序』扫描亦零命中。
- [ok] 五条判据与七步工序的描述前后一致、无矛盾 — 『七步,心里走完,不写进文章』与『五条心内判据…同样不是写出来给人看的』表述一致：过程内化、成品只留骨架与图。
- [ok] ASCII 硬约束可执行 — SKILL.md 明确允许字符集并列禁用/替换对照表（Unicode 箭头/方框/圆点/粗体 → ASCII），约束具体可判定。
- [ok] 外部理论引用属实且非自创 — reach / hard to vary 明确署名 Deutsch《无穷的开始》，作为外部判据引用，非 skill 自创概念。
- [discrepancy] 风险扫描漏报 date 命令与写盘 — §3「執行外部程序」「寫盤」均标 0 命中，但输出节明确要求运行 `date +%Y%m%dT%H%M%S` 并写入 ~/Documents/notes/…__rank.org；扫描为线索非结论，故 file_writes 与原样输出据正文补记。
- [unlocatable] references/template.org 未展开但已定位 — §1 清单列出 skills/ljg-rank/references/template.org（121 bytes，标注 binary），bundle 未给内容（未标『已截断』）；故仅作存在性记入 internal_assets，不据其内容推断行为。
- [ok] meta 直采项无可定位 per-skill 值 — §0 未提供 stars/last_push，installs 无 per-skill 可定位值，按契约均保持 null。
6Verdict & Scenarios
399c5116cb6a5775…