Fewer words, sharper thought. Think twice, write once. Say only what silence cannot carry. The rest, the reader already knows.
npx claudepluginhub gabrielbarberini/laconicBrief prose. Short common words. Trust context. State what matters. Omit what reader can infer.
GabrielBarberini
Philip II: "If I bring my army into your land, I will destroy your farms, slay your people, and raze your city."
Sparta: "If."
Install • Benchmarks • Examples • Caveman vs Laconic • Why
A Claude Code skill that cuts ~75% of tokens — short words, trusted context, implication over explanation.
Inspired by caveman. Same compression, different style.
Say the minimum. Trust context. Every word earns its place.
Real token counts from Claude Sonnet 4 via Langdock API. Each prompt sent with a generic system prompt ("You are a helpful assistant") vs the full laconic SKILL.md. Median of 3 trials per mode.
| Task | Normal (tokens) | Laconic (tokens) | Saved | vs Caveman |
|---|---|---|---|---|
| Explain React re-render bug | 941 | 106 | 89% | +2pp |
| Fix auth middleware token expiry | 1100 | 83 | 92% | +9pp |
| Set up PostgreSQL connection pool | 2104 | 361 | 83% | −1pp |
| Explain git rebase vs merge | 1004 | 347 | 65% | +7pp |
| Refactor callback to async/await | 461 | 141 | 69% | +47pp |
| Architecture: microservices vs monolith | 889 | 356 | 60% | +30pp |
| Review PR for security issues | 758 | 244 | 68% | +27pp |
| Docker multi-stage build | 2583 | 379 | 85% | +13pp |
| Debug PostgreSQL race condition | 1339 | 197 | 85% | +4pp |
| Implement React error boundary | 4096 | 658 | 84% | −3pp |
| Average | 1528 | 287 | 78% | +14pp |
Range: 60%–92% savings across prompts. Caveman benchmarks averaged 64%.
[!IMPORTANT] Laconic only affects output tokens — thinking/reasoning tokens are untouched. Compression targets prose, not cognition. Biggest win is readability and speed; cost savings are a bonus.
cd benchmarks
pip install -r requirements.txt
# Add your Langdock API key to .env.local at the repo root
echo "LANGDOCK_API_KEY=sk-your-key" > ../.env.local
# Dry run (no API calls)
python run.py --dry-run
# Full run (60 API calls: 10 prompts x 2 modes x 3 trials)
python run.py
# Auto-update README table
python run.py --update-readme
🗣️ Normal (69 tokens)
|
🪨 Caveman (19 tokens)
|
Λ Laconic (10 tokens)
|
🗣️ Normal
|
🪨 Caveman
|
Λ Laconic
|
npx skills add GabrielBarberini/laconic
claude plugin marketplace add GabrielBarberini/laconic
claude plugin install laconic@laconic
Install from the .agents/plugins/marketplace.json registry or copy the plugins/laconic/ directory into your Codex plugins folder.
Trigger with:
/laconic