caveman
Ultra-compressed communication mode that cuts ~75% of token use by dropping articles, filler words, and pleasantries while preserving technical accuracy. Use when: long sessions approaching context limits, cost-sensitive API usage, user requests brevity, caveman mode, less tokens, talk like caveman.
Best use case
caveman is best used when you need a repeatable AI agent workflow instead of a one-off prompt.
Ultra-compressed communication mode that cuts ~75% of token use by dropping articles, filler words, and pleasantries while preserving technical accuracy. Use when: long sessions approaching context limits, cost-sensitive API usage, user requests brevity, caveman mode, less tokens, talk like caveman.
Teams using caveman should expect a more consistent output, faster repeated execution, less prompt rewriting.
When to use this skill
- You want a reusable workflow that can be run more than once with consistent structure.
When not to use this skill
- You only need a quick one-off answer and do not need a reusable workflow.
- You cannot install or maintain the underlying files, dependencies, or repository context.
Installation
Claude Code / Cursor / Codex
Manual Installation
- Download SKILL.md from GitHub
- Place it in
.claude/skills/caveman/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How caveman Compares
| Feature / Agent | caveman | Standard Approach |
|---|---|---|
| Platform Support | Not specified | Limited / Varies |
| Context Awareness | High | Baseline |
| Installation Complexity | Unknown | N/A |
Frequently Asked Questions
What does this skill do?
Ultra-compressed communication mode that cuts ~75% of token use by dropping articles, filler words, and pleasantries while preserving technical accuracy. Use when: long sessions approaching context limits, cost-sensitive API usage, user requests brevity, caveman mode, less tokens, talk like caveman.
Where can I find the source code?
You can find the source code on GitHub using the link provided at the top of the page.
SKILL.md Source
# Caveman ## One-Liner Maximum information density, minimum token cost — drop all filler, keep all technical precision. --- ## § 1 · Trigger Phrases Activate when the user says any of: - "caveman mode" - "talk like caveman" - "use caveman" - "less tokens" - "be brief" - `/caveman` Deactivate only on: "stop caveman", "normal mode", "turn off caveman" Once active, **stay active** across the entire session. Do not revert after several turns. --- ## § 2 · Rules **Drop:** - Articles: a, an, the - Filler: just, really, basically, essentially, actually, certainly, sure - Pleasantries: "Great question!", "I'd be happy to...", "Certainly!", "Of course" - Redundant connectors: "In order to", "It is important to note that" - Passive voice when active is shorter **Keep:** - All technical terms, exact and unchanged - Code blocks (code is never compressed) - Numbers and measurements (exact) - Negations (never contract "not" out of existence) **Use:** - Fragments: "Fixed. Push now?" not "I have fixed the issue. Would you like me to push?" - Short words: "big" not "extensive"; "use" not "utilize"; "find" not "discover" - Abbreviations for established terms: DB, auth, config, req/res, fn, impl, spec - Pattern: `[thing] [action] [reason]. [next step].` --- ## § 3 · Compression Examples | Normal | Caveman | |--------|---------| | "I would be happy to help you debug this issue." | "Debug. Go." | | "It is important to note that this function returns null." | "Returns null." | | "I have analyzed the code and found three potential issues." | "Found 3 issues." | | "In order to fix this, you will need to update the configuration." | "Fix: update config." | | "The database query is taking longer than expected." | "DB query slow." | --- ## § 4 · Exceptions **Temporarily exit caveman for:** - Security warnings (precision critical) - Irreversible action confirmations (must be unambiguous) - Multi-step sequences where order matters (numbered steps must be clear) Resume caveman immediately after the exception is communicated. --- ## § 5 · When to Use This Skill **Use when:** - Long coding sessions approaching context window limits - API usage where token cost matters - User has explicitly requested compressed output - Rapid back-and-forth iteration where prose overhead slows flow **Do NOT use when:** - Communicating with non-technical stakeholders - Writing documentation or PRDs (prose quality matters) - Explaining concepts to a beginner (clarity beats brevity)
Related Skills
write-skill
Meta-skill for creating high-quality SKILL.md files. Guides requirement gathering, content structure, description authoring (the agent's routing decision), and reference file organization. Use when: authoring a new skill, improving an existing skill's description or structure, reviewing a skill for quality.
zoom-out
Codebase orientation skill: navigate unfamiliar code by ascending abstraction layers to map modules, callers, and domain vocabulary. Use when: first encounter with unknown code, tracing a data flow, understanding module ownership before editing, orienting before a refactor.
to-prd
Converts conversation context into a structured Product Requirements Document (PRD) and publishes it to the project issue tracker. Do NOT interview the user — synthesize what is already known. Use when: a feature has been discussed enough to capture, converting a design conversation into tracked work, pre-sprint planning.
tdd-workflow
Test-driven development workflow using vertical slices (tracer bullets). Enforces behavior-first testing through public interfaces. Use when: writing new features with TDD, red-green-refactor loop, avoiding implementation-coupled tests, incremental feature delivery.
issue-triage
State-machine issue triage workflow for GitHub, Linear, or local issue trackers. Manages category labels (bug, enhancement) and state labels (needs-triage, needs-info, ready-for-agent, ready-for-human, wontfix). Use when: triaging new issues, clearing needs-triage backlog, routing issues to agents vs humans.
debug-diagnose
Structured six-phase debugging workflow centered on building a reliable feedback loop before theorizing. Use when: debugging hard-to-reproduce issues, performance regression, mysterious failures, agent-assisted root cause analysis, systematic bug fixing.
architecture-review
Codebase architecture review using module depth analysis. Surfaces shallow modules, tight coupling, and locality violations. Proposes deepening opportunities. Use when: pre-refactor audit, tech debt assessment, onboarding architecture review, post-feature architectural cleanup.
vault-secrets-expert
HashiCorp Vault expert: KV secrets, dynamic credentials, PKI, auth methods. Use when managing secrets, setting up PKI, or implementing secrets management. Triggers: 'Vault', 'secrets management', 'HashiCorp Vault', 'dynamic credentials', 'PKI'.
nmap-expert
Expert-level Nmap skill for network reconnaissance, port scanning, service detection, and security assessment. Triggers: 'Nmap', '网络扫描', '端口扫描', 'NSE脚本'. Works with: Claude Code, Codex, OpenCode, Cursor, Cline, OpenClaw, Kimi.
metasploit-expert
Expert-level Metasploit Framework skill for penetration testing, exploit development, and post-exploitation operations. Triggers: 'Metasploit', '渗透测试', '红队', '漏洞利用'. Works with: Claude Code, Codex, OpenCode, Cursor, Cline, OpenClaw, Kimi.
gerrit-permission-manager
Expert manager for Gerrit multi-repository and multi-branch permission configurations. Use when working with Gerrit code review permissions, access controls, repository groups, branch-level permissions, or manifest-based multi-repo management. Use when: gerrit, permissions, code-review, access-control, devops.
container-security-expert
Expert-level Container Security skill using Trivy, Snyk, and other tools for vulnerability scanning, compliance checking, and container hardening. Triggers: '容器安全', '漏洞扫描', 'Trivy', 'Docker安全', 'K8s安全'.