We built PDD (Prompt-Driven Development) to shift the source of truth from code to prompts.
What it is
- CLI that treats prompts as first-class artifacts. You write a prompt; PDD generates the code, a minimal usage example, and tests. It can run a verify step that compares program output against the prompt and iterates until it passes.
What’s different
- Prompts stay in sync with code. If you edit code by hand, `update` back-propagates changes into the prompt so the prompt remains authoritative.
- End-to-end “sync” flow: auto-deps → generate → example → crash/verify → test → fix, with cost/attempt budgets.
- Traceability: map a code line back to the prompt section that produced it.
Try it (no signup)
# quickstart
uv tool install pdd-cli # or: pip install pdd-cli
pdd sync hello_world
Who it’s for
- Engineers who want reproducible codegen with tests and examples, not a one-off LLM paste.
- Teams that want prompts under version control and review.
What we’d love feedback on
- Does the “sync” loop feel reliable on your repo?
- Are the generated tests helpful?
- Where does the trace/back-prop flow break or surprise you?
Caveats
- Local mode uses your own API keys; cloud mode adds stronger models. We’re still improving coverage for large mono-repos.
What it is - CLI that treats prompts as first-class artifacts. You write a prompt; PDD generates the code, a minimal usage example, and tests. It can run a verify step that compares program output against the prompt and iterates until it passes.
What’s different - Prompts stay in sync with code. If you edit code by hand, `update` back-propagates changes into the prompt so the prompt remains authoritative. - End-to-end “sync” flow: auto-deps → generate → example → crash/verify → test → fix, with cost/attempt budgets. - Traceability: map a code line back to the prompt section that produced it.
Try it (no signup) # quickstart uv tool install pdd-cli # or: pip install pdd-cli pdd sync hello_world
Who it’s for - Engineers who want reproducible codegen with tests and examples, not a one-off LLM paste. - Teams that want prompts under version control and review.
What we’d love feedback on - Does the “sync” loop feel reliable on your repo? - Are the generated tests helpful? - Where does the trace/back-prop flow break or surprise you?
Caveats - Local mode uses your own API keys; cloud mode adds stronger models. We’re still improving coverage for large mono-repos.
Links - Repo: [https://github.com/promptdriven/pdd] - Whitepaper: [https://www.notion.so/Whitepaper-The-Case-for-Prompt-Driven-...]