Your first install

ariadnev install writes the kit into the providers you choose. Run it once per project, or once globally, and it records exactly what it wrote in a receipt.

Interactive

av install

You pick providers and scope from a list. Nothing is written until you confirm.

Non-interactive

ariadnev install --provider codex,cursor
ariadnev install --provider claude-code --global
ariadnev install --provider opencode --dry-run     # preview, write nothing

Global flags work everywhere: --home <dir>, --cwd <dir>, --dry-run, --yes.

Scope

• Project (default) writes into the current repository — for example .claude/skills/ for Claude Code — and records .ariadnev/receipt.json there.
• Global (--global) writes into your home directory and records ~/.ariadnev/receipt.json.

What lands where

Each provider has its own conventions. Claude Code receives the widest surface: skills, agents, commands, rules, scripts, hooks, and the statusline. Codex receives skills and scripts at the shared agent home and rules as AGENTS.md. Where a target path or format is not verified for a provider, ariadnev skips that artifact and names it in the install summary rather than writing a guess. The complete matrix is in the provider reference, and ariadnev kit install-path <provider> prints it for one provider.

Claude Code hooks are an event contract: installing to claude-code copies the hook files and, after a y/n confirmation, merges their event bindings into settings.

Check it

ariadnev list            # kit contents and per-provider install state
ariadnev doctor          # health-check the install against its receipt

doctor exits 0 healthy, 1 degraded, 2 unhealthy, so CI can gate on it.

Then, in your agent

With Claude Code, try the daily loop: /av:brainstorm <idea> to explore an approach, /av:plan to phase it, /av:cook <plan path> to implement with tests and review built in. /av:scout <question> answers "where does X live"; /av:fix <bug> proves a root cause before touching code. The full list is the skill catalog.