Agent Scaffold CLI¶
Generate runnable AI agent projects from markdown specs. Pick a recipe, a target language, and a framework — the CLI assembles the relevant docs, asks Claude to emit a complete project, validates the response, and writes the files atomically into your destination of choice.
Published on PyPI as agent-scaffold-cli; the content it composes is fetched live from its two sibling repos, so new recipes and adapters reach you without upgrading.
Install¶
curl -fsSL https://raw.githubusercontent.com/jagguvarma15/agent-scaffold/main/install.sh | sh
The one-liner installs the CLI, puts it on your PATH, and offers to store your Anthropic key. Prefer pipx or uv, or want a one-off run without installing? See Installation.
What you get¶
Spec to running project. A recipe is a production-shaped markdown spec, not a code template. The generator reads it together with framework guides and architecture patterns, asks Claude for the complete project, validates the response against a contract (required files, path safety, static lint, build, smoke check), and stages every file to a temp directory before atomically moving it into place — a failed generation leaves your destination untouched.
One command to a running stack. By default new chains straight into provisioning: install dependencies, start docker services, wire credentials, run migrations, seed dev data, launch the frontend, open the browser. The interactive shell keeps the session alive afterwards for /up, /status, /connect, and /down.
Secrets handled properly. A nine-rule hardening model governs every credential the CLI touches: keyring-first storage with a mode-0600 file fallback, no secrets in argv, output redaction, enforced .gitignore, and first-class revocation — each rule locked in by an audit test. See the security model.
Two arms, one engine¶
This CLI is the composition engine of a three-repo toolchain — the other two feed it in parallel:
flowchart LR
B["agent-blueprints<br/>design core"] --> S["agent-scaffold<br/>composition engine"]
D["agent-deployments<br/>verified options"] --> S
S --> R["Running agent"]
- agent-blueprints — the design core: the kernel, framework-agnostic patterns at five levels, and the spec/IR. What the agent is.
- agent-deployments — the verified options: port-typed adapters and production-shaped recipes, indexed by
catalog.yaml. Which options realize each port. - agent-scaffold — this CLI, which composes a selection from both arms into a running project.
Read The ecosystem for how the two arms funnel into /plan and /generate, and Capabilities for the layers, tiers, and bundles you compose from.
Next steps¶
- Quickstart — from API key to a running agent.
- Interactive shell — the recommended way to work.
- CLI commands and shell commands — every command.