How do I start optimus-prime?

Begin by confirming whether you want to reconfigure; optimus-prime will not modify files until you approve at the gates.

What happens to existing rules or CLAUDE.md?

Rules are optional guardrails. CLAUDE.md can be regenerated as part of the setup. Starter-skills may be copied and adapted; existing content is preserved unless you choose to update it.

What is _apply-all.md and should I touch it?

_apply-all.md is a boilerplate default. It should not be touched by optimus-prime; it serves as a baseline for guardrails and should be left intact.

optimus-prime

npx machina-cli add skill Vibe-Builders/claude-prime/optimus-prime --openclaw

Files (1)

SKILL.md

3.6 KB

Ultrathink.

Role

You are a project configurator. Analyze the codebase and configure Claude to work effectively with it.

Context Assessment

Before starting, check:

.claude/rules/ has path-scoped files besides _apply-all.md? → Ask if user wants to reconfigure (_apply-all.md is a boilerplate default and should not be touched)
.claude/project/ exists? → Ask if user wants to reconfigure
./CLAUDE.md exists? → Ask if user wants to regenerate

Wait for user confirmation before proceeding.

Quick Start

Analyze project codebase (patterns, stack, conventions)
Copy matching starter skills from .claude/starter-skills/ to .claude/skills/, adapt generic parts
Create skills for uncovered stacks via /touch-skill
Create .claude/project/ with on-demand references
Identify .claude/rules/ path-scoped guardrails (if any — rules are optional)
Generate ./CLAUDE.md entry point
Delete .claude/starter-skills/ (starters have been processed)

Core Philosophy

Three-layer system — Skills (framework knowledge) + Rules (guardrails, auto-attach) + Project references (on-demand context)
Context-aware placement — Auto-attach only what prevents wrong code; everything else is on-demand
LLM-driven analysis — Claude explores codebase, not scripts
Leverage existing tools — Use /touch-skill + docs-seeker for skill generation

Decision Matrix

Detected	Where	Rule Test
General framework/library	Skill via `/touch-skill`	—
Project-specific constraint (wrong code even with skill)	`.claude/rules/` with `paths:`	"With the relevant skill activated, will code still be wrong without this?" → Yes
Architecture, structure, domain context	`.claude/project/`	—

Rules are optional. "Important" ≠ "must auto-attach." Only create rules for things skills can't cover.

Red flags that something is NOT a rule:

Applies to all files (**/*.swift, **/*.ts) → probably general knowledge, belongs in skill
A skill already teaches this pattern → redundant, don't duplicate
It's a language/framework feature, not a project decision → skill

Output Structure

./CLAUDE.md                       # Entry point
.claude/rules/
└── <name>.md                     # Path-scoped guardrails (auto-attached)
.claude/project/
└── *.md                          # On-demand references (architecture, structure)

Constraints

Rule files must be concise — guardrails only, not documentation
Rules auto-attach: do NOT reference them in CLAUDE.md
Project references go to .claude/project/ and ARE referenced in CLAUDE.md
Show what will be created before creating
Ask for confirmation at gates

References

Reference	Content
analysis-checklist.md	What to look for in projects
claude-md-template.md	CLAUDE.md format guide

Workflows

Full Setup — Complete project setup flow (default workflow)

Templates

templates/CLAUDE.template.md — CLAUDE.md starter

Additional Context (Optional)

<user-context>$ARGUMENTS</user-context>

Source

git clone https://github.com/Vibe-Builders/claude-prime/blob/main/.claude/skills/optimus-prime/SKILL.mdView on GitHub

Overview

Optimus-prime analyzes the codebase to tailor Claude for a given project. It detects conventions and stack, decides which skills to create, and generates CLAUDE.md entry points with project-specific rules.

How This Skill Works

The role acts as a project configurator: it examines the repository for patterns, stack, and conventions, checks for existing .claude/rules, .claude/project, and CLAUDE.md, and then copies matching starter skills from .claude/starter-skills/. It creates or updates project-specific rules and CLAUDE.md, then prompts for confirmation before applying any changes.

When to Use It

Starting a new project and configuring Claude for that repository
Introducing a new tech stack or framework into an existing project
Reconfiguring Claude when project conventions or architecture change
Regenerating CLAUDE.md entry points after major refactors
Preparing to deploy or hand over a repository with project-specific guardrails

Quick Start

Step 1: Analyze the project codebase (patterns, stack, and conventions)
Step 2: Copy matching starter skills from .claude/starter-skills/ to .claude/skills/ and adapt
Step 3: Generate CLAUDE.md entry point, create .claude/project/ if needed, and review before applying

Best Practices

Ask for user confirmation at each gate before making changes
Do not touch _apply-all.md inside .claude/rules (keep boilerplate safe)
Reuse and adapt matching starter-skills from .claude/starter-skills/ rather than building from scratch
Create rules only for gaps that skills cannot cover; avoid over-attachment
Document exactly what will be created and shown to the user before proceeding

Example Use Cases

Configure Claude for a React + Node API project to auto-attach relevant rules and generate CLAUDE.md
Set up Claude for a Django + PostgreSQL backend with REST endpoints and project-specific guardrails
Add a new microservice in a polyrepo and generate per-service CLAUDE.md and rules
Migrate an in-house monorepo to per-project context with .claude/project references
Regenerate CLAUDE.md after major refactor to reflect new conventions and stacks

Frequently Asked Questions

Add this skill to your agents