Start Using AI on an Existing Project

Your project exists. Your code works. Now make AI understand it in 10 minutes — not 10 hours.

Who This Is For

• Developers with a running project who want AI assistance
• Teams adopting AI coding tools on legacy codebases
• Solo developers looking to 10x their velocity on existing work

Prerequisites: Node.js 22+, Git initialized, any CodyMaster-supported AI agent

What You'll Build

• ✅ Structural map your AI agent reads in one shot
• ✅ Working memory that persists across sessions
• ✅ AGENTS.md manifest — single source of truth for AI behavior
• ✅ AI agent that navigates your code like a senior dev

---

The 10-Minute Setup

Minute 1-2: Install CodyMaster

bash <(curl -fsSL https://raw.githubusercontent.com/tody-agent/codymaster/main/install.sh) --all
codymaster  # Interactive setup — select your AI agent

Minute 3-4: Generate the Skeleton Index

# From your project root
bash scripts/index-codebase.sh

What this creates:

.cm/
└── skeleton.md    # ~5K tokens — every function, class, route

Example output (skeleton.md):

## src/api/
- `export async function getUser(id: string): Promise<User>`
- `export async function createOrder(data: OrderInput): APIResponse`
- `export class AuthService { login, logout, refresh }`

## src/components/
- `export function Dashboard(): JSX.Element`
- `export function UserTable({ users, onSort }): JSX.Element`

Your AI agent reads this ONCE and knows your entire architecture.

Minute 5-6: Initialize Working Memory

@[/cm-start] Initialize working memory for this project

Creates:

.cm/
├── skeleton.md         # Code structure (Layer 0)
├── CONTINUITY.md       # Working memory
├── config.yaml         # RARV settings
└── memory/
    ├── learnings.json  # Error patterns
    └── decisions.json  # Architecture decisions

Minute 7-8: Create AGENTS.md

@[/cm-project-bootstrap] Generate AGENTS.md for this existing project

AGENTS.md is the universal manifest that tells any AI agent:

• Project overview and tech stack
• Coding conventions and patterns
• Safety rules (never modify, always confirm)
• File structure and key entry points

Minute 9-10: First AI-Assisted Task

@[/cm-start] Fix the user authentication timeout bug reported in issue #123

The agent now:

1. Reads AGENTS.md → understands project rules
2. Reads skeleton.md → knows where auth code lives
3. Reads CONTINUITY.md → checks for related past learnings
4. Investigates with precision → fixes with TDD

---

What Changes in Your Daily Workflow

Before CodyMaster

You: "Fix the login bug"
AI:  "Can you show me the project structure?"
You:  *shares 20 files*
AI:  "What framework are you using?"
You:  *explains tech stack*
AI:  "Where is the auth code?"
You:  *points to specific files*
...30 minutes of context-setting later...
AI:  "Here's a fix" (untested, may break other things)

After CodyMaster

You: "@[/cm-start] Fix the login bug"
AI:  *reads skeleton.md + CONTINUITY.md in 2 seconds*
AI:  "I see AuthService.login() in src/api/auth.ts. 
      Last session we noted timeout issues with JWT refresh.
      Let me write a test first..."
AI:  *writes test → implements fix → runs quality gate*
...5 minutes later...
AI:  "Fixed and verified. 3 tests added. Zero regressions."

---

Continuous Improvement

Session End Protocol

Every session, CONTINUITY.md auto-updates with:

• Current task state
• Files modified
• Learnings captured (errors + their prevention)

Adding Intelligence Layers

As you use CodyMaster more, add deeper intelligence:

Project Size	Add This	Benefit
Any	skeleton.md (auto)	Instant structure understanding
50+ files	CodeGraph	Caller/callee tracing, impact analysis
200+ files	cm-deep-search + qmd	BM25 semantic search across all files
500+ files	Full Layer 3	Task-specific context packets

---

Skills Involved

Phase	Skill	Effort
Index	cm-codeintell	Auto (3-4s)
Memory	cm-continuity	Auto
Manifest	cm-project-bootstrap	2 min
Work	cm-start → chain	Depends on task

Tips

• Run skeleton index after major changes — it takes <4s, so run it freely
• Review CONTINUITY.md when switching tasks — it's your team's knowledge base
• Don't skip AGENTS.md — without it, every AI session starts cold
• Commit .cm/ to git — lets your whole team share working memory