mirror of
https://github.com/JuliusBrussee/caveman.git
synced 2026-08-11 13:21:09 +02:00
README: add Wenyan mode, skills, and evals
Expand README with new features and docs: add TOC links for Skills and Evals, update intro to mention Wenyan mode/terse commits/reviews and compression tool, and consolidate Windows Codex install/symlink notes into a short note. Add a new Wenyan (文言文) section with levels and example, introduce Caveman Skills (caveman-commit and caveman-review) with usage examples and severity prefixes, and add an Evals section describing the eval harness and commands. Minor wording/ordering tweaks to installation text.
This commit is contained in:
Julius Brussee
@@ -19,13 +19,15 @@
|
||||
<a href="#benchmarks">Benchmarks</a> •
|
||||
<a href="#before--after">Before/After</a> •
|
||||
<a href="#intensity-levels">Intensity Levels</a> •
|
||||
<a href="#caveman-skills">Skills</a> •
|
||||
<a href="#caveman-compress">Compress</a> •
|
||||
<a href="#evals">Evals</a> •
|
||||
<a href="#why">Why</a>
|
||||
</p>
|
||||
|
||||
---
|
||||
|
||||
A [Claude Code](https://docs.anthropic.com/en/docs/claude-code) skill/plugin and Codex plugin that makes agent talk like caveman — cutting **~75% of output tokens** while keeping full technical accuracy. Plus a companion tool that compresses your memory files to cut **~45% of input tokens** every session.
|
||||
A [Claude Code](https://docs.anthropic.com/en/docs/claude-code) skill/plugin and Codex plugin that makes agent talk like caveman — cutting **~75% of output tokens** while keeping full technical accuracy. Now with [文言文 mode](#文言文-wenyan-mode), [terse commits](#caveman-commit), [one-line code reviews](#caveman-review), and a [compression tool](#caveman-compress) that cuts **~45% of input tokens** every session.
|
||||
|
||||
Based on the viral observation that caveman-speak dramatically reduces LLM token usage without losing technical substance. So we made it a one-line install.
|
||||
|
||||
@@ -165,26 +167,11 @@ Codex:
|
||||
4. Search `Caveman`
|
||||
5. Install plugin
|
||||
|
||||
### Windows (VS Code Codex)
|
||||
|
||||
If you are using the VS Code Codex extension on Windows:
|
||||
|
||||
1. Clone this repo locally
|
||||
2. Open the cloned repo in VS Code
|
||||
3. Open Codex Settings → Plugins
|
||||
4. Look under the repo/local marketplace for `Caveman`
|
||||
5. Click `Install`
|
||||
6. Run `Developer: Reload Window`
|
||||
7. Verify `Caveman` appears under installed plugins
|
||||
|
||||
This repo includes a repo-local marketplace entry at `.agents/plugins/marketplace.json`, so the plugin can be discovered directly from the cloned repository without adding personal machine-specific paths to the repo.
|
||||
|
||||
Install once. Use in all sessions after that.
|
||||
|
||||
One rock. That it.
|
||||
|
||||
> [!NOTE]
|
||||
> **Windows users:** This repo uses symlinks. On Windows, enable `git config core.symlinks true` before cloning (requires developer mode or admin). The plugin installer resolves symlinks to real files at install time, so symlinks work transparently after installation.
|
||||
> **Windows Codex users:** Clone repo → VS Code → Codex Settings → Plugins → find `Caveman` under local marketplace → Install → Reload Window. Also enable `git config core.symlinks true` before cloning (requires developer mode or admin).
|
||||
|
||||
Install once. Use in all sessions after that. One rock. That it.
|
||||
|
||||
## Usage
|
||||
|
||||
Trigger with:
|
||||
@@ -204,12 +191,27 @@ Sometimes full caveman too much. Sometimes not enough. Now you pick:
|
||||
| **Lite** | `/caveman lite` or `$caveman lite` | Drop filler, keep grammar. Professional but no fluff |
|
||||
| **Full** | `/caveman full` or `$caveman full` | Default caveman. Drop articles, fragments, full grunt |
|
||||
| **Ultra** | `/caveman ultra` or `$caveman ultra` | Maximum compression. Telegraphic. Abbreviate everything |
|
||||
| **Wenyan-Lite** | `/caveman wenyan-lite` or `$caveman wenyan-lite` | Semi-classical 文言文. Less intense, grammar intact |
|
||||
| **Wenyan-Full** | `/caveman wenyan` or `/caveman wenyan-full` | Full 文言文. Maximum classical terseness |
|
||||
| **Wenyan-Ultra** | `/caveman wenyan-ultra` or `$caveman wenyan-ultra` | Extreme abbreviation, classical Chinese feel |
|
||||
|
||||
Level stick until you change it or session end.
|
||||
|
||||
### 文言文 (Wenyan) Mode
|
||||
|
||||
Caveman go ancient. Classical Chinese literary compression — same technical accuracy, but in the most token-efficient written language humans ever invented.
|
||||
|
||||
| Level | Trigger | What it do |
|
||||
|-------|---------|------------|
|
||||
| **Wenyan-Lite** | `/caveman wenyan-lite` | Semi-classical. Grammar intact, filler gone |
|
||||
| **Wenyan-Full** | `/caveman wenyan` | Full 文言文. Maximum classical terseness |
|
||||
| **Wenyan-Ultra** | `/caveman wenyan-ultra` | Extreme. Ancient scholar on a budget |
|
||||
|
||||
```
|
||||
English: "Your component re-renders because you create a new object reference each render."
|
||||
Caveman: "New object ref each render. Wrap in useMemo."
|
||||
Wenyan: "物出新參照，致重繪。useMemo Wrap之。"
|
||||
```
|
||||
|
||||
Same answer. Different era. Fewer token.
|
||||
|
||||
## What Caveman Do
|
||||
|
||||
| Thing | Caveman Do? |
|
||||
@@ -223,6 +225,58 @@ Level stick until you change it or session end.
|
||||
| Pleasantries | 💀 "Sure I'd be happy to" is dead |
|
||||
| Hedging | 💀 "It might be worth considering" extinct |
|
||||
|
||||
## Caveman Skills
|
||||
|
||||
Caveman not just talk different. Caveman have **tools**:
|
||||
|
||||
### caveman-commit
|
||||
|
||||
Terse commit messages. Conventional Commits format. Subject ≤50 chars. Body only when "why" not obvious.
|
||||
|
||||
```
|
||||
❌ "feat: add a new endpoint to get user profile information from the database"
|
||||
|
||||
✅ feat(api): add GET /users/:id/profile
|
||||
```
|
||||
|
||||
Trigger: `/caveman-commit` or just ask for a commit message.
|
||||
|
||||
### caveman-review
|
||||
|
||||
One-line code review comments. Location, problem, fix. No throat-clearing.
|
||||
|
||||
```
|
||||
❌ "I noticed that on line 42 you're not checking if the user object
|
||||
is null before accessing the email property. This could potentially
|
||||
cause a crash if the user is not found in the database."
|
||||
|
||||
✅ L42: 🔴 bug: user can be null after .find(). Add guard before .email.
|
||||
```
|
||||
|
||||
Severity prefix tell you what matter:
|
||||
- `🔴 bug:` — broken, will cause incident
|
||||
- `🟡 risk:` — works but fragile
|
||||
- `🔵 nit:` — style, author can ignore
|
||||
- `❓ q:` — genuine question
|
||||
|
||||
Trigger: `/caveman-review` or ask for a code review.
|
||||
|
||||
## Evals
|
||||
|
||||
Caveman not just claim 75%. Caveman **prove** it.
|
||||
|
||||
The `evals/` directory has a three-arm eval harness that measures real token compression against a proper control — not just "verbose vs skill" but "terse vs skill". Because comparing caveman to verbose Claude conflate the skill with generic terseness. That cheating. Caveman not cheat.
|
||||
|
||||
```bash
|
||||
# Run the eval (needs claude CLI)
|
||||
uv run python evals/llm_run.py
|
||||
|
||||
# Read results (no API key, runs offline)
|
||||
uv run --with tiktoken python evals/measure.py
|
||||
```
|
||||
|
||||
Snapshots committed to git. CI runs free. Every number change reviewable as diff. Add a skill, add a prompt — harness pick it up automatically.
|
||||
|
||||
## Why
|
||||
|
||||
```
|
||||
|
||||
Reference in New Issue
Block a user