Skip to content

Latest commit

 

History

History
54 lines (42 loc) · 3.25 KB

PDD.md

File metadata and controls

54 lines (42 loc) · 3.25 KB

Prompt-driven development in mathphysics

Registered PDD scopes use nested configuration and architecture files:

  • Python/rotations: rotation conversion module.
  • Rust/qft-symbolics: fourteen bounded arithmetic/operator/Lorentz/CLI units plus the exact sphere unit; manual adoption, full regeneration unverified. See its PDD.md.
  • Rust/physics-contracts: typed metadata contracts and JSON validation CLI.
  • Rust/lattice-gauge: four exact units (torus geometry, Z2 cochains, Z2 enumeration, purity bounds); manual, prompt-first; see its PDD.md.
  • Rust/physics-tools: bounded JSON bridge to existing exact QFT operations.
  • TypeScript/physics-workbench: six bounded browser TS/CSS units, including PDF.js rendering; manually implemented, automated regeneration unverified. See its PDD.md.

Run commands from the applicable unit directory. Initial agent implementation and path/preprocessing checks do not imply a successful full PDD regeneration; inspect each unit's evidence. LaTeX notes, research notebooks, validation oracles, unmapped native library units, rigid_body.py and simulator_server.py remain outside the prompt graph. Do not infer ownership from the presence of a repository-level PDD guide.

Ernest describes desired behavior in ordinary language. Agents preserve that meaning, run pdd intent plan before mutation, and apply the exact approved intent after meaning approval. Approval from the current conversation persists; do not ask him to choose prompt filenames, dev units, story slugs or commands. For future behavior changes, edit the prompt before regenerating managed code. Keep independent acceptance tests and numerical oracles outside generated output.

Before PDD work, read these canonical Monoclaw playbooks (sibling repository):

  • ../Monoclaw/docs/pdd/PDD_INTENT_FIRST_WORKFLOW.md
  • ../Monoclaw/docs/pdd/PDD_NATURAL_LANGUAGE_AGENT_PLAYBOOK.md
  • ../Monoclaw/docs/pdd/PDD_WITH_ANY_AGENT_HARNESS.md
  • ../Monoclaw/docs/pdd/PDD_PROMPT_GRAPH_AND_ARTIFACT_BUNDLES.md

Use Ernest's ../PromptDrivenDevelopment/pdd/.venv/bin/pdd; installed help is authoritative. Run commands from the applicable nested scope, with PDD_LOCAL_ONLY=1 PDD_SKIP_UPDATE_CHECK=1. Local-only disables GitHub access; model calls can still use configured external providers. Do not change provider or authentication configuration to evade a failed run.

From the repository root, run the independent and integration tests with:

PYTHONDONTWRITEBYTECODE=1 PYTHONPATH=Python/rotations .venv/bin/python -m unittest discover -s Python/rotations/tests -v

The existing uv environment supplies SciPy for dynamics verification; the rotation module itself uses only Python's standard library. HTTP tests need a localhost test socket. Native/CAS suites are additional gates when changing shared mathematical behavior, not evidence supplied by the Python suite.

Current adoption evidence and remaining verification: PDD_ADOPTION.md.

Adoption preserves the existing implementation. Configuration and passing characterization tests alone do not prove that a fresh PDD regeneration passes. Record provider failures and synchronization status honestly; never mark a failed intent run as applied by editing its state. Never hand-edit a generated story contract.