LLM policy
LLM mode must be explicit.
Default behavior must not call OpenAI, Anthropic, OpenRouter, Copilot, Ollama, or any other local or remote model provider.
Provider Selection
Future LLM commands should require explicit provider selection:
branchbrief --llm --provider openaibranchbrief --llm --provider anthropicbranchbrief --llm --provider openrouterbranchbrief --llm --provider ollamaData Modes
| Mode | Intended contents |
|---|---|
metadata | Repo, branch, commit messages, file paths, diff stats, risk signals, suggested verification. |
patch | Selected diff hunks. Requires a warning. |
full | Full diff. Requires a stronger warning. |
The default LLM data mode should be metadata.
API Key Disclosure
Before an LLM call, branchbrief should disclose:
- Provider.
- Model.
- Credential source.
- Data mode.
- Whether network access is used.
It must never print secret values.
Privacy Warnings
Patch and full modes may send source code or sensitive data to the selected provider. They should require a clear warning and an explicit confirmation or non-interactive acknowledgement.