Caveman

Caveman is a coding agent skill that reduces output token count by shortening prose responses.

Caveman is a coding agent skill that reduces output token counts by cutting preamble and shortening prose while keeping code, commands, file paths, and error strings intact. It installs as a single rule file or via a skills CLI without requiring an account or binaries. Users can select different verbosity levels such as lite, ultra, full, and three wenyan variants, or revert to normal mode using text commands. JetBrains testing on 86 real coding tasks showed 8.5 percent fewer output tokens with no detectable quality change. The skill runs alongside various coding agents including Claude Code and JetBrains Junie.

Caveman at a glance

Best for: Software engineer
Pricing: Free
Released: April 2026
Category: AI coding tool
Website: docs.caveman.so

Key features

Installs as a single rule file
Requires no account or binaries
Supports multiple verbosity levels
Preserves code and file paths byte for byte
Tracks mode using disk hooks
Reverts to normal mode via text commands

Use it for

Shrinking coding agent output tokens
Configuring agent verbosity levels
Running paired A/B testing on coding tasks

Alternatives to Caveman

Agent Skills — Encodes senior-engineer workflows and quality gates into 24 skills.
jevgrep — Reduces coding agent costs by providing targeted, relevant repository context.
Squeeze — Cuts token usage for coding agents by shrinking verbose tool output while retaining key data.
I Have ADHD — Stops coding agents from burying answers with direct, action-first steps.
Lcc — Cuts token counts significantly without losing exact session details.
Command Code — Repairs open model tool calls and learns developer patterns automatically.
Pi Skills — Extends coding agents with web search, browser automation, and Google tools.
Skill Seekers — Converts 18 data sources into structured AI skills for 22 platforms.
SimpleEnglish — It enforces strict plain English rules to prevent misread instructions.
Stop That Shit — Enforces task boundaries to stop unnecessary AI agent actions.
Tech Leads Agent Skills — Extends AI coding agents with security-scanned, specialized workflows and tools.
Waza — Embeds structured engineering habits into AI agents to prevent generic, imprecise output.

All Caveman alternatives