Skip to content

development.md

Latest commit

 

History

History
55 lines (37 loc) · 1.58 KB

development.md

File metadata and controls

55 lines (37 loc) · 1.58 KB
Raw

Development

Run Tests

Installed environment:

python -m unittest discover -s tests -v

Directly from the checkout:

PYTHONPATH=src python3 -m unittest discover -s tests -v

Validate Skills And Plugin

for skill in skills/*; do
  python3 /path/to/skill-creator/scripts/quick_validate.py "$skill"
done

python3 /path/to/plugin-creator/scripts/validate_plugin.py plugins/branchforge

The repository includes behavioral skill evals in skills/branching-deliberation/evals/evals.json.

Add A Provider

Implement ModelProvider.complete, expose it in provider_from_name, update the CLI choices, and add parser plus end-to-end tests. Preserve the JSON contracts used by the orchestrator.

Add An Objective Evaluator

The highest-value extension is a domain-specific verifier. It should:

  1. receive a candidate and declared invariants;
  2. run deterministic checks outside the model;
  3. return structured evidence with provenance;
  4. distinguish observed results from model interpretations;
  5. prevent a judge from overriding failed hard constraints.

Contributing

Useful contributions include:

  • objective evaluators for coding, research, and architecture;
  • semantic diversity scoring;
  • provider adapters and resilient request handling;
  • resumable event replay;
  • isolated branch workspaces;
  • benchmark datasets comparing single-agent and branching performance;
  • visualization of live and historical search trees.

Keep control-plane decisions deterministic where practical, preserve failed-search evidence, and add tests for every new state transition.