cc-dejavu: Skill for Claude Code

Ingest Codebase - Mental Model Creation

Create a comprehensive mental model that captures how a system actually works - not just what files exist, but the conceptual boundaries, responsibilities, decisions, and connections that shape it.

When to Use

• Starting a new mental model for a codebase
• The existing model feels incomplete or shallow
• Onboarding to understand a system's architecture
• After major changes to verify the model is still accurate

Philosophy

A good domain model answers:

• "What are the core things this system deals with?" → Domains
• "What can this system do?" → Capabilities (with --operates-on connections!)
• "What concerns cut across multiple parts?" → Aspects (with --applies-to connections!)
• "Why was it built this way?" → Decisions

CRITICAL: Every capability MUST have --operates-on linking it to domains. Every aspect MUST have --applies-to. Orphan entities indicate incomplete modeling.

---

Phase 1: Rapid Codebase Survey (Launch 5-10 Agents in Parallel)

Before modeling, understand the codebase structure. Launch parallel agents to explore different aspects simultaneously:

Agent 1: "Read README.md, package.json, and entry points. What problem does this solve? Who are the users?"
Agent 2: "List all top-level directories. What's in each? (src/, lib/, apps/, packages/, etc.)"
Agent 3: "Find configuration files (*.config.*, .env*, etc.). What can be configured?"
Agent 4: "Search for 'export class' or 'export function' in key directories. What are the main abstractions?"
Agent 5: "Look for test directories. What's being tested? This reveals important functionality."
Agent 6: "Search for API routes, CLI commands, or event handlers. What are the entry points?"
Agent 7: "Look for database models, schemas, or types. What data does this system manage?"
Agent 8: "Search for external integrations (HTTP clients, SDKs, third-party services)."

Output: A map of the codebase with preliminary domain candidates.