LIVE
CliffCompaction: A Context-Compaction Method to Cut Costs of Coding Agents22/09/26|FleXray: A Generalist Model for Whole-Body X-ray Segmentation22/09/26|Anthropic unveils Claude Opus 5.522/09/26 · Anthropic|Court filings reveal internal admissions on AI content scraping19/09/26|US Military Avoids Close Call After AI-Generated Intelligence Hallucination18/09/26|Study finds gender bias in GPT models is not reduced but reshaped across generations17/09/26 · OpenAI|Google DeepMind adds secure server-side memory to Private AI Compute23/09/26 · Google DeepMind|Google DeepMind adds text-to-speech to Gemini 3.823/09/26 · Google DeepMind|NVIDIA releases Nemotron 3 Diarization for real-time speaker identification23/09/26 · NVIDIA|OpenAI extends its Daybreak cyber program to Ukraine for civilian defense23/09/26 · OpenAI|OpenAI enlists influencers to burnish its public image23/09/26 · OpenAI|Claude Code Reportedly Reads AGENTS.md Only When Telemetry Is Enabled23/09/26 · Anthropic|CliffCompaction: A Context-Compaction Method to Cut Costs of Coding Agents22/09/26|FleXray: A Generalist Model for Whole-Body X-ray Segmentation22/09/26|Anthropic unveils Claude Opus 5.522/09/26 · Anthropic|Court filings reveal internal admissions on AI content scraping19/09/26|US Military Avoids Close Call After AI-Generated Intelligence Hallucination18/09/26|Study finds gender bias in GPT models is not reduced but reshaped across generations17/09/26 · OpenAI|Google DeepMind adds secure server-side memory to Private AI Compute23/09/26 · Google DeepMind|Google DeepMind adds text-to-speech to Gemini 3.823/09/26 · Google DeepMind|NVIDIA releases Nemotron 3 Diarization for real-time speaker identification23/09/26 · NVIDIA|OpenAI extends its Daybreak cyber program to Ukraine for civilian defense23/09/26 · OpenAI|OpenAI enlists influencers to burnish its public image23/09/26 · OpenAI|Claude Code Reportedly Reads AGENTS.md Only When Telemetry Is Enabled23/09/26 · Anthropic|
SkillNew

Caveman

Skill : A reusable method: instructions, procedures, sometimes scripts or templates · requires claude-code

Makes the agent respond in telegraphic style: fewer pleasantries and filler, code and error messages kept exactly as-is.

Noticeably reduces output tokens and speeds up responses; the skill alone is sufficient, the proxy and companion engine are under the BSL licence.

Overall score
8.8 /10
Security: nothing to flag
Utility
7.0
Reliability
10.0
Security
No caveat on record
Ease
1 click · Beginner

Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.

What it does, concretely

AI coding agents bill for every token they produce and tend to write long, polite, repetitive responses. Caveman solves this by enforcing a telegraphic style on the agent: filler and pleasantries disappear, while code, file paths, and error messages are kept exactly as-is.

In practice, Caveman consists of a skill (a rule file injected into the agent), a proxy that runs locally between the agent and the AI provider to compress what the agent reads (logs, test output, JSON, diffs), and a middleware for developers embedding an LLM in their own code. The skill alone installs with a single command and works with more than thirty different agents.

Caveman is aimed at developers who use coding agents daily and want to reduce their token bill without degrading response quality. A JetBrains test on 86 real tasks measured 8.5% fewer output tokens with the skill alone, with no detectable quality change.

On the other hand, Caveman is not suited to contexts where detailed, well-written responses are expected, such as documentation writing or exchanges with non-developers. The proxy and compression engine are under the BSL-1.1 licence, which limits commercial use without a separate agreement.

What it offers

  • ·Telegraphic style enforced on the agent: removes pleasantries and filler, keeps code and error messages intact.
  • ·Built-in mode commands: /caveman lite for concise-but-polite, /caveman ultra for minimal responses, /caveman wenyan for classical Chinese.
  • ·/caveman-commit command to generate a Conventional Commit in a single line.
  • ·/caveman-review command for condensed code review, one finding per line with line number and severity level.
  • ·/caveman-compress command to rewrite memory files (CLAUDE.md) in telegraphic style with automatic backup of the original.
  • ·Local proxy that compresses agent inputs (logs, JSON, diffs, test output) before each provider call, with originals kept on disk (paid, BSL-1.1).
  • ·TypeScript and Python middleware to integrate compression into your own LLM applications (paid, BSL-1.1).
  • ·caveman learn command to analyse local agent history and rank the biggest token sinks, with a suggested fix for each.

In practice

Price
Free
Commercial use
Commercial use: yes
Available on
GitHub, skills.sh, Claude Code plugin
Host
claude-code
Licence
MIT
Other hosts
claude-code, codex, cursor, gemini-cli
FR interface
No

Security and data

Rating
Security: nothing to flag
Your data
Data stays local
Maintenance
Active
Last release
15 September 2026

Links and repository checked on 23 September 2026 · added on 23 September 2026

How to install Caveman

1. Make sure Node.js 22.13 or higher is installed on your machine. 2. Run: npx skills add JuliusBrussee/caveman -g 3. Open your agent (Claude Code or compatible): the skill activates automatically. 4. If the agent does not respond, type /caveman to activate it manually. 5. To test, ask a coding question and observe the reduction in filler in the response. 6. To uninstall: npx -y github:JuliusBrussee/caveman -- --uninstall
npx skills add JuliusBrussee/caveman -g

Official documentation

In the same category

Whole category Route & save