메뉴
Ultra-compressed communication mode that keeps technical accuracy while stripping filler. Use when the user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes "/caveman".
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Helps analyze systemic problems, teach systems-thinking concepts, and turn conclusions into explainable diagrams, model portfolios, Remotion animation briefs, presentation storyboards, and visual assets. Use when the user wants to understand a complex recurring problem, learn how to read systems diagrams, apply many-model thinking, compare multiple models, prepare a presentable explanation, animate a system map, create a Remotion video plan, make a presentation deck, generate conceptual images, mentions systems thinking, causal loops, feedback loops, Sherwood, Seeing the Forest for the Trees, Scott E. Page, The Model Thinker, stock and flow, leverage points, root causes, unintended consequences, or asks to make a diagram of a business, product, organizational, social, or operational system.
Stress-test a user's plan or design by interviewing them through the decision tree until shared understanding is reached. Use when the user wants to be grilled, stress-test a plan, sharpen a design, or mentions "grill me".
Create portable agent skills with clear triggers, concise instructions, progressive disclosure, and optional bundled resources. Use when the user wants to create, write, build, improve, or package a new skill.
Use when setting up a new project, auditing an existing one, or aligning a project so the same project-level skills and agent guide work under Claude Code, Codex, and Pi Agent. Trigger on phrases like "make this project canonical", "set up canonical skills layout", "ensure canonical skills style", "migrate CLAUDE.md to AGENTS.md", "make claude/codex/pi share the same agent guide here", "standardize the agent layout for this repo", or whenever the user opens a fresh project and wants the five project invariants (skills/ real dir, .claude/skills symlink, .agents/skills symlink, AGENTS.md canonical, CLAUDE.md = @AGENTS.md) to hold. Also trigger when the user mentions only one of these symptoms (drifted CLAUDE.md vs AGENTS.md, .claude/skills or .agents/skills as a real directory, multiple skills directories) — the goal is to converge to the canonical shape every time, not just react to the symptom they named.
AI image generation, video generation, SVG/vector creation, and image editing powered by WaveSpeed AI CLI. Use when the user wants to generate images, create videos from text or images, edit/modify images with AI, create visual assets (hero images, thumbnails, placeholders, social media graphics), animate still images into video, batch-generate multiple assets, iterate on designs, generate SVG/vector logos or icons, restore old photos, relight images, colorize B&W photos, swap faces, or segment images. Also triggers for: 'make me an image', 'generate a thumbnail', 'create a placeholder', 'animate this', 'edit this photo', 'change the background', 'generate assets for my project', 'I need visuals', 'make me a logo', 'restore this old photo', 'relight this', 'which wavespeed model should I use', or any request involving AI-powered image/video creation or editing — even if the user doesn't mention WaveSpeed by name.
Craft high-quality prompts for Google Nano Banana Pro image generation. Use when the user wants to write, improve, or refine an image generation prompt, needs help describing a visual concept for AI generation, wants to adapt a prompt to a different style (photorealistic, anime, 3D, etc.), wants to build a prompt from a reference image description, asks 'how do I prompt for...', 'write me a prompt', 'improve this prompt', 'make this more cinematic/anime/realistic', 'nanobana prompt', 'nano banana prompt', 'I want to make an image of...', 'help me with my image prompt', 'this image gen result looks bad', 'show me example prompts', or any request about prompt engineering for image generation — even if they don't mention Nano Banana by name.
| name | caveman |
| description | Ultra-compressed communication mode that keeps technical accuracy while stripping filler. Use when the user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes "/caveman". |
| license | MIT |
| version | 0.1.0 |
| homepage | https://github.com/mattpocock/skills/tree/main/skills/productivity/caveman |
| compatibility | {"claude-code":"supported","codex":"supported","pi-agent":"supported"} |
| metadata | {"upstream":"mattpocock/skills/skills/productivity/caveman"} |
Respond terse like smart caveman. Keep technical substance. Remove fluff.
Once triggered, stay active every response. Do not drift back after several turns. Turn off only when user says "stop caveman" or "normal mode".
Drop articles, filler, pleasantries, and hedging. Fragments are okay. Prefer short synonyms and common technical abbreviations such as DB, auth, config, req, res, fn, and impl. Use arrows for causality when they help.
Keep technical terms exact. Keep code blocks unchanged. Quote errors exactly.
Preferred pattern:
[thing] [action] [reason]. [next step].
Avoid:
Sure! I'd be happy to help you with that. The issue you're experiencing is likely caused by...
Prefer:
Bug in auth middleware. Token expiry check use `<` not `<=`. Fix:
User: "Why React component re-render?"
Assistant:
Inline obj prop -> new ref -> re-render. `useMemo`.
User: "Explain database connection pooling."
Assistant:
Pool = reuse DB conn. Skip handshake -> fast under load.
Temporarily drop caveman style for security warnings, irreversible action confirmations, multi-step sequences where fragments risk misread, or when the user asks to clarify or repeats a question. Resume after the clear part is done.