Project Init & Auto-Context

Learn how VibeCody’s smart project scanner gives the AI deep understanding of your codebase — automatically, in every conversation.

Prerequisites: VibeCody installed with a working provider. See First Provider if needed.

The Problem

When you ask an AI agent to “fix the tests” or “add auth”, it doesn’t know:

What language or framework your project uses
What the build and test commands are
Where the entry points and config files are
What environment variables are needed

Without this context, the agent guesses — and often guesses wrong.

The Solution: `/init`

VibeCody’s project scanner detects all of this automatically and injects it into every agent conversation.

Step 1: Scan Your Project

vibecli
> /init

Output:

Scanning project...

Project: my-saas-app
  A full-stack SaaS application with auth and billing.
Architecture: full-stack application
Languages: TypeScript, Rust
Frameworks: React, Vite, Axum, Tokio
Package managers: pnpm, cargo

Build commands:
  pnpm run build → pnpm run build
  Cargo build → cargo build --workspace

Test commands:
  Vitest → pnpm run test (Vitest)
  Cargo test → cargo test --workspace (cargo test)

Lint commands:
  ESLint → npx eslint .
  Clippy → cargo clippy --workspace -- -D warnings

Entry points: src/App.tsx, src/main.rs
Expected env vars: DATABASE_URL, STRIPE_SECRET_KEY, JWT_SECRET

Key files: README.md (readme), Dockerfile (Dockerfile), .github/workflows/ci.yml (CI config)

Yes Project profile cached to .vibecli/project-profile.json
   This context will be auto-injected into every agent session.

Step 2: Use the Agent — It Already Knows Your Project

Now when you run an agent task, VibeCody automatically injects the project profile:

> /agent add rate limiting to the API

The agent now knows:

Your API is built with Axum (Rust)
Tests run with cargo test
The entry point is src/main.rs
It should use cargo clippy to verify after changes

Without /init, the agent would have to spend multiple steps figuring this out.

What Gets Detected

Category	Detected For
Languages	Rust, TypeScript/JavaScript, Python, Go, Java, C#, Ruby, PHP
Frameworks	React, Next.js, Vue, Svelte, Angular, Express, FastAPI, Django, Flask, Axum, Gin, Rails, Laravel, and 15+ more
Build tools	cargo, npm/pnpm/yarn/bun, go, maven, gradle, dotnet, bundler, composer
Test frameworks	cargo test, Jest, Vitest, Playwright, Cypress, pytest, go test, JUnit, PHPUnit, RSpec
Lint tools	Clippy, ESLint, Prettier, Biome, Ruff, go vet
Architecture	Monorepo, library, CLI tool, full-stack app, microservice cluster, single package
Key files	README, Dockerfile, CI config, API specs, env examples, test configs, schemas
Env vars	Extracted from `.env.example` / `.env.sample` / `.env.template`

Task-Based Auto-Context

When you give the agent a task, VibeCody also analyzes the task description and auto-gathers relevant files:

Task mentions…	Auto-gathered files
“test”, “spec”	`tests/`, `jest.config.js`, `pytest.ini`
“build”, “compile”	`Cargo.toml`, `package.json`, `tsconfig.json`
“deploy”, “CI”	`Dockerfile`, `.github/workflows/`, `docker-compose.yml`
“auth”, “login”	`src/auth/`, `src/middleware/`
“database”, “schema”	`prisma/`, `migrations/`, `src/models/`
“API”, “endpoint”	`src/api/`, `src/routes/`, `pages/api/`
“style”, “CSS”	`tailwind.config.js`, `src/styles/`

Up to 5 relevant files are previewed and injected into the agent’s context window.

How It Works Under the Hood

/init calls project_init::scan_workspace() which reads package manifests, config files, and directory structure
The profile is cached to .vibecli/project-profile.json (1-hour TTL)
On every agent run, run_agent_repl_with_context() loads the cached profile
build_system_prompt() injects the profile summary into the LLM system prompt
extract_relevant_files_for_task() analyzes the task and gathers file previews

The profile is re-scanned automatically if the cache is older than 1 hour.

AI Panel Context (VibeUI)

In VibeUI, the same project context is available:

@project in the chat context picker injects the project profile
ProjectContextPanel shows the full profile with 4 tabs: Overview, Commands, Key Files, AI Context
One-click build/test/lint execution from the Commands tab

Combining with `/orient`

For deeper analysis, use /orient after /init:

> /orient

This sends the auto-detected profile to the AI and asks it to analyze:

Architecture patterns and design decisions
Recommended next steps for a new developer
Potential improvements and technical debt
CI/CD recommendations

/orient uses the cached profile as a starting point, so it’s faster and more accurate than running without /init.

Tips

Run /init once when you open a new project — it caches for 1 hour
The .vibecli/project-profile.json file is auto-generated; add it to .gitignore
For monorepos, /init detects the workspace structure and lists all members
The profile works with all 23 providers — the context is injected at the prompt level

Tutorial: Project Init & Auto-Context