{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreiahjs6t5shx55yqiucy5aztge2zow7s4mk2jc47xs4rvvktdriwma",
    "uri": "at://did:plc:5opbpi2nomj4y3d5kpwamkrd/app.bsky.feed.post/3mmvkcuvj3gu2"
  },
  "coverImage": {
    "$type": "blob",
    "ref": {
      "$link": "bafkreiabnsczlffnj6dp4an4wxramuuechkcz5ub6lfvk62y4d3c7sjtxe"
    },
    "mimeType": "image/png",
    "size": 322518
  },
  "description": "Large Language Models are incredibly powerful for software engineering, but they also have a habit of being verbose. Long explanations, conversational filler, and repeated context all consume tokens, increase latency, and dilute the signal-to-noise ratio during AI-assisted engineering.\n\nThe “caveman” skill for GitHub Copilot CLI takes the opposite approach: aggressively concise communication while preserving the technical substance.\n\nInstead of:\n\n“Sure! I’d be happy to help you debug that issue.",
  "path": "/install-the-caveman-skill-for-github-copilot-cli-system-wide/",
  "publishedAt": "2026-05-28T07:56:35.000Z",
  "site": "https://corti.com",
  "tags": [
    "GitHub Copilot CLI Documentation",
    "Copilot CLI page",
    "JuliusBrussee/caveman"
  ],
  "textContent": "Large Language Models are incredibly powerful for software engineering, but they also have a habit of being verbose. Long explanations, conversational filler, and repeated context all consume tokens, increase latency, and dilute the signal-to-noise ratio during AI-assisted engineering.\n\nThe “caveman” skill for GitHub Copilot CLI takes the opposite approach: aggressively concise communication while preserving the technical substance.\n\nInstead of:\n\n> “Sure! I’d be happy to help you debug that issue. It looks like there may be a problem in your authentication middleware…”\n\nYou get:\n\n> “Bug in auth middleware. Token null after refresh. Fix session propagation.”\n\nMinimal words. Maximum information density.\n\nThis post explains how to install the caveman skill system-wide for GitHub Copilot CLI and why this style can materially improve AI-assisted development workflows.\n\n* * *\n\n# What Is the Caveman Skill?\n\nThe caveman skill modifies the communication style of GitHub Copilot CLI responses to make them:\n\n  * Extremely terse\n  * Technically dense\n  * Low-noise\n  * Token efficient\n\n\n\nThe style intentionally removes:\n\n  * Pleasantries\n  * Hedging\n  * Filler words\n  * Excess explanation\n  * Conversational overhead\n\n\n\nWhile preserving:\n\n  * Technical accuracy\n  * Code\n  * Commands\n  * Important warnings\n  * Critical reasoning\n\n\n\nThe result feels closer to reading optimized engineering notes than chatting with a traditional assistant.\n\n* * *\n\n# Why Developers Like This Style\n\n## 1. Reduced Token Usage\n\nLLM context windows are finite resources.\n\nVerbose responses waste:\n\n  * Prompt tokens\n  * Completion tokens\n  * Context budget\n  * Attention\n\n\n\nA concise interaction style means:\n\n  * More room for actual code\n  * Larger repositories fit into context\n  * Longer agentic sessions before truncation\n  * Lower API costs in some scenarios\n\n\n\nThis becomes especially important during:\n\n  * Repo-scale engineering\n  * Agentic coding workflows\n  * Multi-step debugging sessions\n  * Long Copilot CLI conversations\n\n\n\n* * *\n\n## 2. Better Signal-to-Noise Ratio\n\nTraditional assistant responses often contain conversational padding:\n\n  * “I’d be happy to help”\n  * “It seems like”\n  * “You may want to consider”\n  * “One possible solution is”\n\n\n\nExperienced developers usually do not need this.\n\nCaveman mode compresses output into:\n\n\n    Root cause: race condition in cache invalidation.\n    Fix lock ordering. Add retry.\n\n\nThe important information becomes immediately visible.\n\n* * *\n\n## 3. Faster Cognitive Parsing\n\nEngineering work already overloads working memory:\n\n  * Terminal output\n  * Stack traces\n  * Logs\n  * Diff reviews\n  * Infrastructure configs\n\n\n\nShorter AI responses reduce cognitive switching costs.\n\nInstead of reading paragraphs, developers scan concise technical fragments.\n\nThis works particularly well in:\n\n  * Terminal-based workflows\n  * SSH sessions\n  * Remote debugging\n  * Pair-programming with AI\n  * Fast iteration loops\n\n\n\n* * *\n\n## 4. Better Fit for Agentic Engineering\n\nModern AI-assisted engineering increasingly relies on:\n\n  * Autonomous agents\n  * Iterative execution\n  * Small-step task loops\n  * Continuous verification\n\n\n\nIn these workflows, verbose natural language becomes friction.\n\nConcise responses improve:\n\n  * State tracking\n  * Action chaining\n  * Context preservation\n  * Tool orchestration\n  * Agent memory efficiency\n\n\n\nThis aligns well with modern approaches such as:\n\n  * Spec-driven development\n  * AI-assisted repo maintenance\n  * Continuous validation loops\n  * Multi-agent engineering systems\n\n\n\n* * *\n\n# Install GitHub Copilot CLI\n\nBefore installing the caveman skill, install GitHub Copilot CLI.\n\nSee the official documentation at:\n\nGitHub Copilot CLI Documentation and the Copilot CLI page\n\nAuthenticate and verify functionality first.\n\nExample:\n\n\n    gh copilot suggest \"find largest files\"\n\n\nIt also works in the Copilot Chat interface\n\n\n    copilot\n\n* * *\n\n# Install the Caveman Skill System-Wide\n\nRun the following command:\n\n\n    cd ~ && npx -y github:JuliusBrussee/caveman -- --only copilot\n\n\nThis installs the caveman integration for GitHub Copilot CLI into your home directory configuration.\n\nThe repository is available here:\n\nJuliusBrussee/caveman\n\n* * *\n\n# Create Global Copilot Instructions\n\nCreate the file:\n\n\n    ~/.copilot/copilot-instructions.md\n\n\nAdd the following content:\n\n\n    Respond terse like smart caveman. All technical substance stay. Only fluff die.\n\n    Rules:\n    - Drop: articles (a/an/the), filler (just/really/basically), pleasantries, hedging\n    - Fragments OK. Short synonyms. Technical terms exact. Code unchanged.\n    - Pattern: [thing] [action] [reason]. [next step].\n    - Not: \"Sure! I'd be happy to help you with that.\"\n    - Yes: \"Bug in auth middleware. Fix:\"\n\n    Switch level: /caveman lite|full|ultra|wenyan\n    Stop: \"stop caveman\" or \"normal mode\"\n\n    Auto-Clarity: drop caveman for security warnings, irreversible actions, user confused. Resume after.\n\n    Boundaries: code/commits/PRs written normal.\n\n\nThis enables the behavior globally for GitHub Copilot CLI.\n\n* * *\n\n# Verify Configuration\n\nTest with:\n\n\n    gh copilot suggest \"why docker container exits immediately\"\n\n\nTypical normal output:\n\n\n    Container likely exiting because main process terminates immediately.\n    Check ENTRYPOINT and CMD configuration.\n\n\nTypical caveman output:\n\n\n    Main process die. Container exit.\n    Check ENTRYPOINT/CMD.\n\n\nSame meaning. Fewer tokens.\n\n* * *\n\n# Caveman Modes\n\nThe configuration supports multiple intensity levels:\n\n## Lite\n\nSlightly compressed responses.\n\nGood balance between readability and efficiency.\n\n\n    Cache invalidation bug. Refresh stale.\n\n\n* * *\n\n## Full\n\nAggressive compression.\n\n\n    Cache stale. Invalidate after write.\n\n\n* * *\n\n## Ultra\n\nMaximum terseness.\n\n\n    Cache stale. Flush.\n\n\n* * *\n\n## Wenyan\n\nExtremely condensed style inspired by classical Chinese brevity.\n\nMostly novelty/fun mode.\n\n* * *\n\n# When Caveman Mode Automatically Disables\n\nThe configuration intentionally drops the caveman style during situations where clarity matters more than brevity:\n\n  * Security warnings\n  * Destructive operations\n  * Irreversible actions\n  * Potential user confusion\n\n\n\nThis is important because excessive terseness can become dangerous during:\n\n  * Production infrastructure changes\n  * Database deletions\n  * Credential management\n  * Security incident handling\n\n\n\nThe configuration resumes terse mode afterward.\n\n* * *\n\n# Why This Matters for AI-Assisted Engineering\n\nThe industry trend is moving toward:\n\n  * AI agents\n  * Continuous tool orchestration\n  * Large-context workflows\n  * Autonomous repo reasoning\n  * Long-running coding sessions\n\n\n\nIn these environments, verbosity becomes operational overhead.\n\nConcise prompting and concise responses improve:\n\nArea| Benefit\n---|---\nContext Window| More usable space\nToken Cost| Lower consumption\nLatency| Faster responses\nReadability| Faster scanning\nAgentic Workflows| Better chaining\nCognitive Load| Reduced fatigue\n\nThis mirrors traditional engineering optimization principles:\n\n  * Reduce unnecessary state\n  * Compress signal\n  * Remove redundancy\n  * Preserve essential information\n\n\n\nCaveman mode applies those principles to human-AI interaction itself.\n\n* * *\n\n# Example Workflow\n\nNormal style:\n\n\n    I think the issue may be related to your Kubernetes readiness probe configuration. The container appears to be starting correctly, but the readiness check may be failing before the application fully initializes.\n\n\nCaveman style:\n\n\n    Readiness probe fail before app ready.\n    Increase initialDelaySeconds.\n\n\nFor experienced engineers, the second version is often enough.\n\n* * *\n\n# Caveats\n\nCaveman mode is not ideal for every scenario.\n\nLess suitable for:\n\n  * Junior developers\n  * Teaching\n  * Architecture discussions\n  * Documentation writing\n  * Complex design rationale\n  * Cross-team communication\n\n\n\nBest use cases:\n\n  * Fast debugging\n  * CLI workflows\n  * DevOps tasks\n  * Iterative coding\n  * AI pair programming\n  * Terminal-heavy environments\n\n\n\nThe ideal workflow is often hybrid:\n\n  * Caveman for rapid iteration\n  * Normal mode for final explanations and documentation\n\n\n\n* * *\n\n# Final Thoughts\n\nMost AI UX optimization focuses on improving the model.\n\nCaveman mode optimizes something different:\n\n> Communication entropy.\n\nFor experienced developers, removing conversational overhead can make AI tooling feel dramatically faster, sharper, and more aligned with terminal-centric engineering workflows.\n\nAs AI-assisted engineering evolves toward persistent agents and large-context automation, concise interaction styles may become increasingly valuable—not just stylistically, but operationally.",
  "title": "Install the “Caveman” Skill for GitHub Copilot CLI System-Wide",
  "updatedAt": "2026-05-28T07:56:35.768Z"
}