An interactive wizard that audits your repo and generates the Claude Code config it needs.
Caution
Not yet released. See the issue tracker for status.
agentboost is a single command that opens an interactive wizard, drives a Claude agent loop to audit your repository, and writes the config files it produces.
- Detects an existing
CLAUDE.mdor.claude/and asks whether to refresh, replace, merge, or cancel - A read-only agent explores your repo using
list_dir,read_file,grep, andgit_log— no write or shell access - Each generated artifact lands in a tabbed preview where you accept, skip, or mark for regeneration
- Regeneration re-invokes the agent loop with scope narrowed to just that artifact's kind
- Writes are atomic (tempfile + rename) and only happen after you confirm
brew install aniftyco/tap/agentboostagentboost # launch the wizard in the current directory
agentboost --help # full flag listInside the wizard:
↑/↓move the cursorspacetoggles a scope option←/→switch artifact tabs in reviewa/s/rset the per-artifact action (accept / skip / regenerate)enterconfirms the current screenesccancels — no files are written
agentboost needs an Anthropic API key. Set ANTHROPIC_API_KEY in your environment, or drop a config file at ~/.config/agentboost/config.toml:
[provider]
default = "anthropic"
[provider.anthropic]
api_key = "sk-ant-..."
model = "claude-opus-4-7"
max_tokens = 8192
[agent]
max_iterations = 40Precedence: CLI flag > env var > config file > built-in default.
v0.1 ships Claude (Anthropic API) as the only provider and Claude Code as the only output target. The internals split along two axes — ModelProvider (which LLM does the thinking) and target (which agent's config format you generate) — so additional providers (OpenAI, Ollama, Claude CLI shell-out) and targets (Cursor, Aider, and so on) slot in cleanly later.
- macOS or Linux
- An Anthropic API key
agentboost is open-source software licensed under the MIT license.