# Caveman Lite, Full, Ultra, and Wenyan Intensity Levels Explained > Understand Caveman intensity levels: lite, full, ultra, and wenyan. Discover how each level minimizes tokens while preserving accuracy for efficient code documentation. - Repository: [Julius Brussee/caveman](https://github.com/JuliusBrussee/caveman) - Tags: deep-dive - Published: 2026-07-07 --- **Caveman provides four core intensity levels—lite, full, ultra, and wenyan—that progressively strip filler words, grammatical articles, and linguistic flourishes to minimize token count while preserving technical accuracy and code integrity.** The **Caveman** project by JuliusBrussee is a token-compression mode that rewrites AI responses in a terse, caveman-like style. According to the repository's configuration in [`skills/caveman/SKILL.md`](https://github.com/JuliusBrussee/caveman/blob/main/skills/caveman/SKILL.md), users can select from six distinct intensity variations (including wenyan sub-levels) to control exactly how aggressively the system removes superfluous language. ## Understanding the Four Core Intensity Levels Each intensity level applies a predefined rule set that governs exactly which tokens are removed or transformed. The definitions are baked into the skill's configuration file, with each tier offering a different balance between brevity and readability. ### Lite Mode: Conservative Filler Removal **Lite** mode removes only obvious filler words (e.g., "just", "really") and hedging phrases while keeping articles and full grammatical sentences. The prose remains professional but compact, offering a modest token reduction that preserves most natural-language cues. This level is ideal when you want cleaner output without sacrificing readability. ### Full Mode: The Default Caveman Style **Full** is the default intensity that activates when you invoke the mode without specifying a level. This level drops articles completely, permits sentence fragments, and substitutes longer synonyms with shorter alternatives. It also eliminates tool-call narration, decorative tables, and emojis, yielding the classic "caveman" tone with substantial token savings. ### Ultra Mode: Maximum Compression **Ultra** strips conjunctions whenever meaning remains clear, collapses multi-word expressions to single words, and forbids any prose abbreviations (e.g., `cfg`, `impl`). This level removes arrows and invented shortcuts, providing maximum token reduction while strictly preserving code symbols and exact strings. Use this when you need the absolute minimum character count for technical responses. ### Wenyan Modes: Classical Chinese Translation The **wenyan** levels translate responses into Classical Chinese (文言文) rather than compressing English text. The implementation includes three sub-levels: **wenyan-lite** retains classical register with basic grammar, **wenyan-full** compresses to 80-90% fewer characters using classical syntax, and **wenyan-ultra** pushes compression to the absolute limit while maintaining the classical feel. These modes provide language-specific compression optimized for character-based efficiency. ## Architectural Implementation Details The intensity system relies on several core mechanisms defined in the source code to ensure consistent behavior across conversations. ### Persistence Across Sessions The selected intensity persists across all subsequent replies until a new `/caveman` command or a "stop caveman" trigger overrides it. This **Persistence** behavior is enforced by the configuration block in [`skills/caveman/SKILL.md`](https://github.com/JuliusBrussee/caveman/blob/main/skills/caveman/SKILL.md), ensuring your chosen compression level remains active without requiring re-entry. ### Rule Engine Processing Each intensity level swaps in a specific rule set that governs token removal algorithms. For example, the **lite** rule only drops filler phrases, while the **ultra** rule also removes conjunctions and prohibits invented abbreviations. These rule definitions live in the "Rules" section of [`skills/caveman/SKILL.md`](https://github.com/JuliusBrussee/caveman/blob/main/skills/caveman/SKILL.md). ### Auto-Clarity Safety Guard The mode automatically falls back to normal language when compression would jeopardize safety or create ambiguity. The **Auto-Clarity Guard** described in [`SKILL.md`](https://github.com/JuliusBrussee/caveman/blob/main/SKILL.md) specifically protects security warnings and critical technical instructions from being over-compressed. ### Command Interface Users switch levels using the slash command `/caveman `, documented in [`commands/caveman.md`](https://github.com/JuliusBrussee/caveman/blob/main/commands/caveman.md). The interface accepts any of the six levels: `lite`, `full`, `ultra`, `wenyan-lite`, `wenyan-full`, or `wenyan-ultra`. ## Practical Usage Examples The following examples demonstrate how the same technical explanation transforms under different intensity levels. Switch to lite mode: ```text /caveman lite ``` Expected output: ```text Your component re‑renders because you create a new object reference each render. Wrap it in `useMemo`. ``` Switch to full mode: ```text /caveman full ``` Expected output: ```text New object ref each render. Inline object prop = new ref = re‑render. Wrap in `useMemo`. ``` Switch to ultra mode: ```text /caveman ultra ``` Expected output: ```text Inline obj prop, new ref, re‑render. `useMemo`. ``` Switch to wenyan-full mode: ```text /caveman wenyan-full ``` Expected output: ```text 每繪新生對象參照，故重繪；以 useMemo 包之則免。 ``` ## Summary - **Caveman** offers four core intensity levels—**lite**, **full**, **ultra**, and **wenyan**—each defined in [`skills/caveman/SKILL.md`](https://github.com/JuliusBrussee/caveman/blob/main/skills/caveman/SKILL.md). - **Lite** removes only filler words; **full** drops articles and narration; **ultra** forbids abbreviations and conjunctions; **wenyan** translates to Classical Chinese with three sub-levels. - The **Persistence** mechanism maintains your selected level across replies until explicitly changed. - **Auto-Clarity Guard** automatically disables compression for safety-critical content to prevent ambiguity. - Use `/caveman ` (documented in [`commands/caveman.md`](https://github.com/JuliusBrussee/caveman/blob/main/commands/caveman.md)) to switch between modes, with **full** being the default. ## Frequently Asked Questions ### What is the default Caveman intensity level? The default intensity is **full**, as defined in [`skills/caveman/SKILL.md`](https://github.com/JuliusBrussee/caveman/blob/main/skills/caveman/SKILL.md) and referenced in [`src/rules/caveman-openclaw-bootstrap.md`](https://github.com/JuliusBrussee/caveman/blob/main/src/rules/caveman-openclaw-bootstrap.md). This level drops articles, permits sentence fragments, and eliminates decorative elements while maintaining technical clarity. You can override this default at any time using the `/caveman` command. ### How does the wenyan intensity level differ from standard compression? While **lite**, **full**, and **ultra** operate in English by removing tokens, the **wenyan** levels translate responses into Classical Chinese (文言文) with increasing terseness. **Wenyan-lite** retains basic classical grammar, **wenyan-full** compresses character count by 80-90%, and **wenyan-ultra** pushes compression to the limit while preserving the classical register. ### Can I switch intensity levels mid-conversation? Yes, the system uses a **Persistence** mechanism defined in [`skills/caveman/SKILL.md`](https://github.com/JuliusBrussee/caveman/blob/main/skills/caveman/SKILL.md) that maintains your selected intensity across all subsequent replies. Simply issue a new `/caveman ` command to change modes, or use the "stop caveman" trigger to disable the mode entirely. ### Will Caveman ever disable compression automatically? Yes, the **Auto-Clarity Guard** automatically falls back to normal language when compression would jeopardize safety or create ambiguity, such as with security warnings. This safeguard ensures that critical technical warnings remain fully comprehensible regardless of the selected intensity level.