REPL reference

agent-scaffold scaffold opens the shell. Input is either a slash command, a bare recipe slug (demo means /recipe demo), or free text interpreted as a plan refinement. Completion is fuzzy; unknown commands get a "did you mean" hint.

Session

Command	Purpose
/help	List available commands. /help refine lists free-text refinement keys.
/config	Set up credentials: the Anthropic key + any env vars the stack needs.
/reset	Drop the current draft. Keeps config + resolved sources, clears everything else.
/draft	List, save, resume, or delete a named selection draft (at most 3 kept).
/exit	Leave the REPL.

Compose the plan

Command	Purpose
/new	Guided wizard: recipe → language → framework → name → dest → optional features → plan.
/recipe <slug>	Select the recipe. Bare /recipe lists slugs; a partial query filters.
/language <lang>	Pick the target language (python or typescript).
/framework <name>	Pick the framework. Validated against the recipe's declared dependencies.
/tier [T0..T4]	Show or set the capability tier (/tier clear resets).
/layer <layer> <ids...>	Inspect or set one layer's capabilities (e.g. /layer memory cache.redis vector_db.qdrant).
/observability <backend> [cloud\|docker]	Pick the observability backend and where it runs.
/stack [<layer>\|<id>]	Browse every stack option in the catalog, grouped by layer; /stack <id> shows a detail card.
/name <project>	Set the project name (auto-derives /dest if not set).
/dest <path>	Override the destination directory.
/model <id>	Override the model id (e.g. /model claude-sonnet-5).
/effort low\|medium\|high	Apply an effort preset (model + max_tokens + thinking + strict).

Inspect

Command	Purpose
/plan	Render the generation plan + cost with the current selections.
/context	Show the full context-tier breakdown plus dropped / truncated lists.
/status	Readiness check: Anthropic key, Docker, and the selected stack's env vars.
/sync	Re-sync deployments + blueprints from GitHub and reload recipes.

Generate and run

Command	Purpose
/generate	Confirm + run the generation pipeline (the final step of /new).
/autorun on\|off	Whether /generate chains into up + welcome panel + browser open.
/docker on\|off\|auto	Run mode for /up and autorun: containers, local, or auto. Never affects generation.
/write_mode <mode>	How /generate handles existing files in dest (abort, skip, diff, overwrite).
/open <path>	Attach the session to an existing generated project.
/up	Bring the generated project's stack up (docker sandbox / local servers).
/down	Tear down the local stack: stop the servers + docker compose down.
/connect [<option>]	Connect a stack option (docker or cloud hosted) — runs in the REPL.

Aliases

Alias	Resolves to
/quit, /q	/exit
/h, /?	/help
/go, /gen	/generate
/write-mode	/write_mode
/load	/open
/drafts	/draft

Deprecated

These still dispatch for one release with a migration hint, then disappear: /deploy (use agent-scaffold deploy --target <t>), /eval (use agent-scaffold eval --cwd <dest>), /logs (use agent-scaffold logs <service>), /cost (cost is part of /plan).

Free-text refinement keys

Anything that isn't a slash command is interpreted by a small Haiku call into a typed patch over the session. The accepted keys:

Key	Meaning
model	Override model (e.g. claude-sonnet-5, claude-haiku-4-5, claude-opus-4-8).
effort	Preset bundle: low / medium / high (model + tokens + thinking + strict).
framework	Framework name (e.g. langgraph, pydantic_ai, vercel_ai_sdk).
language	Target language: python / typescript.
strict	Toggle the strict generation prompt.
max_tokens	Anthropic max_tokens cap for this run.
thinking_budget	Extended-thinking token budget (null disables).
stack_mode	Capability stack mode: quick / customize.
tier	Capability tier T0..T4; none clears.
rag_preset	RAG preset: simple / complex (expands to catalog capability bundles).
add_dependencies	Extra pins to inject: {language: {package: version}}.
add_steps	Extra post-write steps to run (e.g. [docker_up, seed]).
remove_steps	Post-write steps to skip (e.g. [smoke_test]).
remove_roles	Multi-agent roles to drop.
add_capabilities	Capability ids to enable (e.g. [obs.langfuse]).
remove_capabilities	Capability ids to drop.
hosting_overrides	Hosting per capability: {capability id: cloud \| docker}.
notes	Free-form guidance appended verbatim to the LLM prompt.