Summary
Add an LLM Council agent to hve-core that sends the same question package to GPT-5.4, Opus 4.6, and Gemini 3.1 Pro councilor subagents, then synthesizes consensus, disagreement, assumptions, and a recommended next step into one user-facing answer.
Rationale
hve-core already offers structured research, planning, and review workflows. A council pattern complements those workflows for ambiguous or high-judgment tasks by making agreement and disagreement across models explicit instead of relying on a single answer.
Scope
- Add the top-level LLM Council orchestrator agent.
- Add the GPT-5.4, Opus 4.6, and Gemini 3.1 Pro councilor subagents.
- Keep the feature focused on dispatch, independent responses, and synthesis.
- Update collection and generated plugin documentation so the capability is discoverable.
Acceptance Criteria
- hve-core includes an LLM Council agent.
- The agent dispatches the same question package to the three councilor subagents.
- Each councilor responds independently without trying to converge with the others.
- The final synthesis clearly distinguishes consensus, disagreement, uncertainty, and recommendation.
- The agent degrades clearly when one councilor is unavailable.
- Collection metadata and generated plugin documentation surface the new capability.
Summary
Add an LLM Council agent to hve-core that sends the same question package to GPT-5.4, Opus 4.6, and Gemini 3.1 Pro councilor subagents, then synthesizes consensus, disagreement, assumptions, and a recommended next step into one user-facing answer.
Rationale
hve-core already offers structured research, planning, and review workflows. A council pattern complements those workflows for ambiguous or high-judgment tasks by making agreement and disagreement across models explicit instead of relying on a single answer.
Scope
Acceptance Criteria