Spaces:
Running on Zero
Running on Zero
| # HearthNet Agent Coordination | |
| ## Active Roles | |
| | Agent | Role | Ownership | | |
| | --- | --- | --- | | |
| | Codex lead | Integration, docs, final verification, deployment | Whole repo coordination | | |
| | Planck | Phase 1 doc synthesis | Read-only docs review | | |
| | Avicenna | Prototype assessment | Read-only HTML/README review | | |
| | Kepler | HF Space and quality checklist | Read-only deployment/tooling review | | |
| | Mill | Python Phase 1 core | `hearthnet/`, `tests/` | | |
| | Anscombe | Gradio Space UI | `app.py`, optional `static/` | | |
| | Hypatia | Quality/tool config | `pyproject.toml`, requirements, config files | | |
| | Darwin | M01-M07 + contract/glossary audit | Read-only coverage review | | |
| | Gibbs | M08-M13 audit | Read-only coverage review | | |
| | Franklin | X01-X04 + overview/PRD audit | Read-only coverage review | | |
| | Mendel | No-mock local-first model architecture audit | Read-only implementation review | | |
| | Pascal | Gradio/server/UI audit | Read-only UI/runtime review | | |
| | Erdos | Coordination docs policy audit | Read-only tasks/agents review | | |
| ## Collaboration Rules | |
| - Workers have disjoint write scopes. | |
| - No worker may revert another worker's edits. | |
| - Services do not import each other directly; they communicate through the bus. | |
| - UI talks to the controller/facades, not directly to services. | |
| - No mocks or fake AI paths in implementation-facing code. Phase 1 may keep clearly labeled prototype/demo surfaces, but shipped services must use real local-first components or explicit unavailable/degraded states. | |
| - Local AI must be local-first: prefer Ollama, llama.cpp, or local Hugging Face Transformers backends. OpenAI may be used only as an opt-in online fallback when local models are unavailable or explicitly disabled. | |
| - Do not add security-tool suppression pragmas, broad ignores, or Bandit/Ruff/Pylint bypasses to pass checks. Fix the finding or document a narrow, reviewed exception in `tasks.md`. | |
| - Quality gates must be run before deployment. | |
| - UI must follow the spec architecture: UI talks through controller/facades/bus snapshots only, with no direct service imports. | |
| - Spec adherence is a quality gate: changes must map to the relevant M/X docs, capability contract, and glossary terms. | |
| ## Integration Checklist | |
| - [ ] Merge worker changes. | |
| - [ ] Resolve conflicts without losing existing prototypes. | |
| - [ ] Update `tasks.md` statuses. | |
| - [ ] Verify no mocks, fake model responses, or unlabeled simulations remain in implementation paths. | |
| - [ ] Verify local-first model backends are real and OpenAI is only an opt-in online fallback. | |
| - [ ] Verify no new security pragmas, blanket ignores, or quality bypasses were introduced. | |
| - [ ] Verify UI behavior and wording do not overclaim missing spec features. | |
| - [ ] Verify implemented behavior is traceable to M01-M13, X01-X04, `CAPABILITY_CONTRACT.md`, and `GLOSSARY.md`. | |
| - [ ] Run all requested checks. | |
| - [ ] Commit and push to the Hugging Face Space. | |