Skip to content

caveman-spellbook.md

Latest commit

 

History

History
191 lines (135 loc) · 8.56 KB

caveman-spellbook.md

File metadata and controls

191 lines (135 loc) · 8.56 KB
Raw
title The Caveman Spellbook
description Six grunt-level compression modes for agent-almanac content — a homage to JuliusBrussee/caveman, from lite filler-stripping to extreme classical Chinese abbreviation
category reference
agents
translator
project-manager
skill-reviewer
librarian
teams
caveman-spellbook
skills
translate-content

The Caveman Spellbook

Every agent-almanac skill carries words that perform rather than inform. The caveman-spellbook strips those words — across six increasingly radical compression lenses — until only substance remains. It is a homage to JuliusBrussee/caveman, a vocabulary compression framework that challenged how much of written communication is load-bearing.

This guide explains the six levels, shows side-by-side examples drawn from the canonical heal reference translations, and explains how to invoke the caveman-spellbook team for bulk translation waves.

When to Use This Guide

  • To understand what the six grunt levels mean and how they differ
  • To read agent-almanac skills in your preferred cognitive density
  • To activate the caveman-spellbook team for bulk translation of new skills
  • To contribute new canonical reference translations and extend the pilot

The Six Levels

Locale dir Level Style summary
caveman-lite Lite Remove filler/hedging. Keep grammar, articles, full sentences. Professional but tight.
caveman Full Drop articles too. Fragments OK. Classic grunt pattern: [thing] [action] [reason].
caveman-ultra Ultra Abbreviations (DB/auth/config/req/res/fn/impl), causality arrows (→), one word when sufficient.
wenyan-lite Wenyan Lite Semi-classical Chinese. Drop modern filler (其實/基本上). Keep grammar structure.
wenyan Wenyan Full Full 文言文. Classical sentence patterns, verb-object order, subjects often omitted.
wenyan-ultra Wenyan Ultra Extreme ancient compression. Single characters where possible. Maximum terseness.

Side-by-Side Examples

All examples drawn from heal — the canonical reference skill translated in all six levels. Source: skills/heal/SKILL.md.

Section heading: ## Common Pitfalls

Level Rendered heading
English (source) ## Common Pitfalls
caveman-lite ## Pitfalls
caveman ## Pitfalls
caveman-ultra ## Traps
wenyan-lite ## 常見陷阱
wenyan ## 陷
wenyan-ultra ## 忌

Pitfall text: "Performative self-assessment"

English source:

Performative self-assessment: Going through the motions without honest evaluation produces no value. The point is to find real drift, not to demonstrate the ability to self-reflect

caveman-lite:

Performative self-assessment: Going through the motions without honest evaluation produces no value. The point is to find real drift, not to demonstrate the ability to self-reflect

caveman:

Performative self-assessment: Going through motions without honest evaluation. Point is real drift, not demonstrating ability to self-reflect

caveman-ultra:

Performative assessment: Motions ≠ value. Real drift matters.

wenyan-lite:

表演性自我評估：走過場而沒有誠實評估不產生任何價值。目標是發現真實偏移，而非展示自我反思能力

wenyan:

表演性自察：走過場無誠實評估不生任何價值。目的在察真實偏移，非展示自省

wenyan-ultra:

表演性察：走過場→無價值

Marker pair: Expected: / On failure:

Level Expected marker On failure marker
English **Expected:** **On failure:**
caveman-lite **Got:** **If fail:**
caveman **Got:** **If fail:**
caveman-ultra → If err:
wenyan-lite **預期：** **失敗時：**
wenyan 得： 敗則：
wenyan-ultra 得： 敗：

When to Use entry: "Mid-session fatigue"

English source:

Mid-session fatigue: responses feel formulaic, repetitive, or disconnected from the user's actual needs

caveman-lite:

Mid-session fatigue: responses feel formulaic, repetitive, or disconnected from the user's needs

caveman:

Mid-session fatigue: responses formulaic, repetitive, disconnected from user needs

caveman-ultra:

Responses formulaic/repetitive → mid-session fatigue

wenyan-lite:

會話疲勞：回應公式化、重複，或與用戶需求脫節

wenyan:

應答陳套，與用者需求脫節乃用

wenyan-ultra:

應答陳套→用

Style Rules

What always survives

At every compression level, these elements are never altered:

  • Code blocks (```, bash, yaml, r, etc.)
  • Technical identifiers (function names, file paths, CLI flags)
  • Tool names (Read, Write, Edit, Bash, Grep, Glob)
  • YAML frontmatter field names
  • Checklist items (- [ ])
  • Warnings for destructive or irreversible operations (auto-clarity rule)

Auto-clarity rule

Both caveman and wenyan levels revert to clear prose for:

  • Security warnings
  • Destructive action confirmations
  • Multi-step sequences where fragment order risks misread

The compressed style resumes immediately after the clear part.

Glossaries

Style rules are codified in two machine-readable glossaries:

  • i18n/glossaries/caveman.yml — drop patterns, abbreviation table, section heading mappings per level
  • i18n/glossaries/wenyan.yml — classical particles, sentence patterns, per-level compression targets

Reading the Spectrum

To see the full compression spectrum for heal:

# English source
cat skills/heal/SKILL.md

# Six compressed versions
for locale in caveman-lite caveman caveman-ultra wenyan-lite wenyan wenyan-ultra; do
  echo "=== $locale ==="
  cat i18n/$locale/skills/heal/SKILL.md
done

Spot-check the wenyan-ultra version for heal — it compresses the 232-line English source to under 200 lines while retaining all six procedure steps, all validation items, and all pitfall entries.

Running the Team

The caveman-spellbook team coordinates bulk translation across all six locales in two parallel waves:

Activate the caveman-spellbook team for skills: <skill-1>, <skill-2>.
Each translator: read your glossary and heal canonical reference, scaffold
both skills for your locale, translate following your level's compression rules.
Translator frontmatter field: "Julius Brussee homage — caveman".

Wave 1 (caveman trio) and Wave 2 (wenyan trio) run in parallel. After both complete, skill-reviewer does a cross-level QA pass, then librarian runs npm run translation:status.

Scale-up path

The pilot covers 3 skills (heal canonical + commit-changes + make-fire). To add more:

  1. Re-activate caveman-spellbook with the new skill list
  2. Each translator scaffolds and translates their locale
  3. skill-reviewer verifies substance preservation
  4. librarian runs npm run translation:status

Troubleshooting

wenyan-ultra loses technical substance: Classical compression can drop load-bearing verbs. Have skill-reviewer verify all 6 procedure steps appear, even if they are single characters. The rule: every English step number must have a corresponding step at every level.

Code blocks translated: The translator violated the auto-clarity rule. Every code block must be identical to the English source. Fix by copying the source code block verbatim.

caveman-ultra = caveman (no distinction): The translator did not apply abbreviations or causality arrows. Check i18n/glossaries/caveman.yml ultra abbreviations table and re-translate with explicit contractions.

wenyan-lite reads like modern Chinese: The level target is classical register, not just simplified prose. Add classical particles (之/乃/為/其/而/於/以) at sentence boundaries and remove modern filler markers listed in the glossary.

Related Resources