Repository navigation
The Team
create_app_from_scratch.py generates a small Python application from a description in natural language, using several LLM roles rather than one prompt: planning, backend, frontend, database schema, tests, documentation, security review, and refinement passes over what the earlier roles produced.
Full documentation is in docs/theteam.md, which covers installation, usage, the TOML configuration, the generated project structure, custom prompts, the internal logic and its limitations. This page is the short version.
The roles are prompts under prompts/, and reading them is the fastest way to understand what the tool will do:
| Stage | Prompts |
|---|---|
| Planning |
prompt_plan.txt, prompt_description.txt
|
| Backend |
prompt_backend.txt, prompt_backend_refine.txt
|
| Frontend |
prompt_frontend.txt, prompt_frontend_refine.txt
|
| Database |
prompt_database_schema.txt, prompt_database_schema_refine.txt, prompt_data_samples.txt
|
| Quality |
prompt_tests.txt, prompt_documentation.txt, prompt_security_review.txt, prompt_add_security.txt
|
| Assets | prompt_design_asset.txt |
The _refine prompts are second passes: the model is shown its own output and asked to improve it. That is where most of the quality difference comes from, and also most of the token cost.
Prototyping. It produces a starting point for an application you then take over, not something to deploy. The upstream documentation lists its limitations explicitly, and the project's own status note calls the whole toolkit experimental.
Two things follow from that:
-
Read the generated code before running it. In particular anything touching a database, the filesystem, or the network. A
prompt_security_review.txtstage exists, and a model reviewing its own output is not an audit - Generate into an empty directory. It writes a project tree, and you want a clean boundary between what it produced and what you wrote
Same idea as the Optimizer: a TOML file plus command-line options, and an OpenAI-compatible endpoint that can be a local model. See docs/theteam.md for the keys, and Configuration and prompts for the config-file caveat that applies across this repository.