claude-prime
CLAUDE.md119 starsChanged 6 months agoMITView on GitHub
This file provides guidance to Claude Code when working with this repository. A boilerplate toolkit that supercharges Claude Code with skills, commands, and workflows for agent-skill-driven development. Context engineering is about curating the optimal set of tokens to maximize desired model behavior. Unlike prompt engineering (crafting perfect prompts), context engineering considers the holistic state available to the LLM at any moment. Sources: Anthropic - Effective Context Engineering, Manus - Context Engineering Lessons Template skill kits shipped with the prime repo.…
In this repository22 other files it gives its agents.
- agent-browserSkill
- askSkill
- cookSkill
- create-docSkill
- diagnoseSkill
- discussSkill
- docs-seekerSkill
- fixSkill
- frontend-designSkill
- give-planSkill
- media-processorSkill
- optimus-primeSkill
- prime-syncSkill
- review-codeSkill
- self-evolveSkill
- skill-creatorSkill
- testSkill
- backend-fastapi-pythonSkill
- dockerSkill
- frontend-developmentSkill
- humanizerSkill
- monorepoSkill
The file
We split this CLAUDE.md into the rules it gives an agent
3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 36
Don't load everything upfront.
- Distinctive: line 42
Use filesystem as extended memory.
- Distinctive: line 52
Minimal overlap: If engineers can't definitively choose the right tool, agents won't either
≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
CLAUDE.md
This file provides guidance to Claude Code when working with this repository.
Project overview
A boilerplate toolkit that supercharges Claude Code with skills, commands, and workflows for agent-skill-driven development.
Context engineering philosophy
Context engineering is about curating the optimal set of tokens to maximize desired model behavior. Unlike prompt engineering (crafting perfect prompts), context engineering considers the holistic state available to the LLM at any moment.
Core Principles
1. Context is a Finite Resource
Models have limited attention budgets that degrade as context grows (context rot). Goal: find the smallest high-signal set of tokens that maximizes desired outcomes. Every token competes for attention.
2. Right-Altitude Prompting
Concrete but not hardcoded. Find the Goldilocks zone.
| Too Low (Brittle) | Too Vague | Right Altitude |
| "Use Tailwind's blue-500" | "Make it look good" | "Use project's existing color system" |
| "Set font-size to 16px" | "Use appropriate sizing" | "Match surrounding components" |
3. Avoid Railroading Claude
Claude will generally try to stick to your instructions, and because Skills are so reusable you’ll want to be careful of being too specific in your instructions. Give Claude the information it needs, but give it the flexibility to adapt to the situation
4. Don’t State the Obvious
Claude Code knows a lot about your codebase, and Claude knows a lot about coding, including many default opinions. If you’re publishing a skill that is primarily about knowledge, try to focus on information that pushes Claude out of its normal way of thinking.
5. Anti-Convergence Patterns
LLMs default to high-probability outputs. Counter by listing what to avoid, suggesting uncommon alternatives, varying examples, and breaking monotony.
6. Just-in-Time Context Retrieval
Don't load everything upfront. Maintain lightweight identifiers (file paths, URLs) and retrieve dynamically at runtime.
7. Sub-Agent Architecture
Specialized sub-agents handle focused tasks with clean context windows, returning condensed summaries. Prevents context pollution, enables long-horizon tasks.
8. Structured External Memory
Use filesystem as extended memory. Write persistent notes/plans outside context window. Maintain todo lists in recent attention span.
9. Error Preservation
Keep failures visible in context. Failed attempts inform better strategies.
Tool Design Principles
- Self-contained: Each tool independently understandable
- Minimal overlap: If engineers can't definitively choose the right tool, agents won't either
- Descriptive parameters: Unambiguous input names and descriptions
- Token efficient: Minimize wasted exploration
Sources: Anthropic - Effective Context Engineering, Manus - Context Engineering Lessons
System architecture
Skills (.claude/skills/)
| Type | When to create | Examples |
| Workflow skill | Repeatable process with clear actions and instructions | /give-plan, /cook, /fix, /test,/discuss |
| Domain knowledge skill | Domain has project-specific patterns the agent must follow | Created per-project during priming or via /skill-creator |
Starter Skills (.claude/starter-skills/)
Template skill kits shipped with the prime repo. During priming, optimus-prime copies relevant starters into the target project's .claude/skills/, adapts generic parts, and deletes the starter-skills/ folder.
Hub-and-spoke pattern: Complex skills use SKILL.md as the hub (always loaded, <500 lines) that dispatches to spoke files (references/, workflows/, templates/) loaded on demand. This keeps the main file focused while supporting deep reference content.
Knowledge system
| Layer | Location | Loading | Content |
| CLAUDE.md | CLAUDE.md | Always | Entry point — project identity, commands, stack, on-demand reference pointers |
| Skills | .claude/skills/ | On-demand by agent | Framework/library knowledge, workflows |
| Rules | .claude/rules/ | Auto-attached by path | Guardrails — wrong code if missed |
| Agent memory | .claude/agent-memory-local/ | Auto-injected per agent | Runtime-discovered knowledge |
Rule test: "With the relevant skill activated, will an agent still produce incorrect code without this?" Yes -> rule. No -> skill. Rules are optional.
Memory test: "Can this only be learned by working in the project, not authored upfront?" Yes -> agent memory. No -> skill or rule.
On-demand references can live anywhere — just point to them from CLAUDE.md. Use docs/ for human documentation. For complex projects where human docs are too verbose for agents, .claude/project/ can optionally hold agent-optimized references (dense, scannable format).
Configuration scope
| Put here | When |
CLAUDE.md | Team-shared instructions checked into source control — architecture, coding standards, common workflows |
CLAUDE.local.md | Personal preferences for this project (gitignored) — your role, sandbox URLs, preferred test data, workflow quirks |
.claude/rules/_apply-all.md | Universal rules for all agents (main + subagents) — including interaction guardrails |
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.