avc init

Initialize AVC for a project. Creates .avc/ with a SQLite database, default config, and .avcignore.

If the project is inside a git repository and has no .gitignore yet, AVC creates one with its entries; if a .gitignore already exists, AVC appends .avc/ and .avcignore to it.

Usage

avc init                            # initialize current directory
avc init /path/to/project           # initialize a specific directory
avc init --skills claude-code       # also write agent integration files
avc init --skills claude-code,cursor,windsurf

Flags

Flag	Description
--skills <list>	Comma-separated agent frameworks to set up. Supported: claude-code, cursor, windsurf, generic
--yes, -y	Skip the confirmation prompt when no AVC project exists at the path yet
--json	JSON output

Confirmation prompt

If no AVC project exists at the target path yet, avc init asks before creating one:

⚠ No AVC project found at /path/to/project
  Initialize a new AVC project here? [y/N]

Answering anything other than y/yes aborts with no changes made. This only fires on first-time bootstrap — re-running avc init on an already-initialized project never prompts.

Pass --yes to skip the prompt in scripts or CI. --json mode also skips it, since machine consumers are presumed to know what they’re asking for.

JSON output

{
  "id": "proj-a1b2c3",
  "path": "/path/to/project",
  "name": "project",
  "already_initialized": false,
  "success": true
}

What it creates

.avc/
├── avc.db
├── config.toml
├── .gitignore
└── objects/
.avcignore                    ← in the project root

Re-running on an already-initialized project is a no-op — no data loss.

With --skills

For each requested framework, AVC writes two kinds of file: the MCP server config (so the framework knows how to launch avc mcp serve) and instruction files (so the agent knows when to use each tool).

Where the framework supports it, the MCP config is written in the project, so the AVC server is scoped to that project rather than registered globally for every project on your machine. Frameworks without a project-level config option fall back to their home-directory config:

Framework	MCP config	Instruction files (in the project)
claude-code	.mcp.json (project)	CLAUDE.md, .claude/skills/avc-*/SKILL.md
cursor	.cursor/mcp.json (project)	.cursor/rules/avc.mdc
windsurf	~/.codeium/windsurf/mcp_config.json (home)	.windsurfrules
generic	—	AGENT_INSTRUCTIONS.md

Every file AVC creates is added to .gitignore automatically — generated agent files are local tooling, not repo content. Files you already own are left as-is: if a CLAUDE.md or MCP config already exists, AVC appends to (or merges into) it and does not gitignore it, so your tracking choice stands. To share an AVC-generated config, remove its entry from .gitignore and commit it.

Re-running --skills is safe: existing files are never overwritten, the JSON config is merged (not duplicated), and rules files are append-only with a dedup marker. See Agent Integration for details.