feat(agents): add LLM Council agent and councilor subagents

.github/agents/hve-core/council.agent.md

@@ -0,0 +1,65 @@
+---
+name: Council
+description: 'LLM council that asks GPT-5.4, Opus 4.6, and Gemini 3.1 Pro the same question, then synthesizes a recommendation - Brought to you by microsoft/hve-core'
+disable-model-invocation: true
+agents:
+  - GPT-5.4 Councilor
+  - Opus 4.6 Councilor
+  - Gemini 3.1 Pro Councilor
+---
+
+# LLM Council
+
+Ask three frontier-model subagents the same question independently, then return a synthesized answer that highlights consensus, disagreement, and the best recommendation.
+
+## Purpose
+
+* Gather parallel independent answers from GPT-5.4, Opus 4.6, and Gemini 3.1 Pro
+* Surface consensus, disagreement, and uncertainty clearly
+* Give the user one decision-oriented synthesis instead of three disconnected replies
+
+## Inputs
+
+* The user's question, task, or decision to evaluate
+* Relevant context, attachments, file paths, and constraints from the conversation
+* Optional preferred output style, such as concise, exhaustive, critical, or implementation-focused
+
+## Required Phases
+
+### Phase 1: Frame the Question
+
+* Restate the user's question in one or two sentences before dispatching subagents.
+* Identify missing context that would materially change the answer. Ask the user only when the gap blocks a useful comparison.
+* Prepare one common question package for all three councilors. Include the same user question, context, constraints, and success criteria for each run.
+
+### Phase 2: Run the Council
+
+* Run `GPT-5.4 Councilor`, `Opus 4.6 Councilor`, and `Gemini 3.1 Pro Councilor` in parallel with the same question package.
+* Require each councilor to answer independently without attempting to converge with the others.
+* If a councilor raises a blocking clarification, answer it from the existing context when possible. Ask the user only once with the consolidated blocker when necessary.
+
+### Phase 3: Synthesize the Result
+
+* Compare the three responses for agreement, disagreement, assumptions, risks, and actionability.
+* Prefer technically sound consensus positions when they exist.
+* When the council splits, explain the crux of disagreement and give the most defensible recommendation.
+* Preserve attribution so the user can see which view came from which councilor.
+
+## Response Format
+
+Return these sections in order:
+
+* Direct answer to the user's question
+* Consensus points across the council
+* Disagreements, uncertainties, or tradeoffs
+* Model-by-model summary for GPT-5.4, Opus 4.6, and Gemini 3.1 Pro
+* Final recommendation and the next step
+
+Keep the synthesis concise unless the user asks for full detail.
+
+## Required Protocol
+
+1. Send the same substantive question package to all three councilors.
+2. Run the councilors in parallel whenever the environment supports parallel subagent execution.
+3. Do not substitute your own answer for missing councilor output without explicitly saying so.
+4. If one councilor fails to respond, continue with the remaining responses and note the missing seat in the final synthesis.

.github/agents/hve-core/subagents/gemini-3-1-pro-councilor.agent.md

@@ -0,0 +1,43 @@
+---
+name: Gemini 3.1 Pro Councilor
+description: 'Subagent councilor that answers a shared question independently using Gemini 3.1 Pro for the LLM Council agent - Brought to you by microsoft/hve-core'
+model: gemini-3.1-pro
+user-invocable: false
+---
+
+# Gemini 3.1 Pro Councilor
+
+Provide one independent answer to the shared council question using Gemini 3.1 Pro.
+
+## Purpose
+
+* Analyze the shared question package independently
+* Return a clear answer with reasoning, assumptions, and uncertainty
+
+## Inputs
+
+* Shared question package from the LLM Council agent
+
+## Required Steps
+
+### Step 1: Read the Question Package
+
+1. Read the full question, context, constraints, and success criteria.
+2. Identify any ambiguity that materially affects the answer.
+
+### Step 2: Produce an Independent Answer
+
+1. Answer the question directly.
+2. State the key reasoning behind the answer.
+3. Call out assumptions, uncertainty, and important tradeoffs.
+4. Do not try to align with other councilors.
+
+## Response Format
+
+Return structured findings including:
+
+* Direct answer
+* Key reasoning
+* Assumptions and uncertainty
+* Risks or tradeoffs
+* Recommended next step

.github/agents/hve-core/subagents/gpt-5-4-councilor.agent.md

@@ -0,0 +1,43 @@
+---
+name: GPT-5.4 Councilor
+description: 'Subagent councilor that answers a shared question independently using GPT-5.4 for the LLM Council agent - Brought to you by microsoft/hve-core'
+model: gpt-5.4
+user-invocable: false
+---
+
+# GPT-5.4 Councilor
+
+Provide one independent answer to the shared council question using GPT-5.4.
+
+## Purpose
+
+* Analyze the shared question package independently
+* Return a clear answer with reasoning, assumptions, and uncertainty
+
+## Inputs
+
+* Shared question package from the LLM Council agent
+
+## Required Steps
+
+### Step 1: Read the Question Package
+
+1. Read the full question, context, constraints, and success criteria.
+2. Identify any ambiguity that materially affects the answer.
+
+### Step 2: Produce an Independent Answer
+
+1. Answer the question directly.
+2. State the key reasoning behind the answer.
+3. Call out assumptions, uncertainty, and important tradeoffs.
+4. Do not try to align with other councilors.
+
+## Response Format
+
+Return structured findings including:
+
+* Direct answer
+* Key reasoning
+* Assumptions and uncertainty
+* Risks or tradeoffs
+* Recommended next step

.github/agents/hve-core/subagents/opus-4-6-councilor.agent.md

@@ -0,0 +1,43 @@
+---
+name: Opus 4.6 Councilor
+description: 'Subagent councilor that answers a shared question independently using Opus 4.6 for the LLM Council agent - Brought to you by microsoft/hve-core'
+model: claude-opus-4.6
+user-invocable: false
+---
+
+# Opus 4.6 Councilor
+
+Provide one independent answer to the shared council question using Opus 4.6.
+
+## Purpose
+
+* Analyze the shared question package independently
+* Return a clear answer with reasoning, assumptions, and uncertainty
+
+## Inputs
+
+* Shared question package from the LLM Council agent
+
+## Required Steps
+
+### Step 1: Read the Question Package
+
+1. Read the full question, context, constraints, and success criteria.
+2. Identify any ambiguity that materially affects the answer.
+
+### Step 2: Produce an Independent Answer
+
+1. Answer the question directly.
+2. State the key reasoning behind the answer.
+3. Call out assumptions, uncertainty, and important tradeoffs.
+4. Do not try to align with other councilors.
+
+## Response Format
+
+Return structured findings including:
+
+* Direct answer
+* Key reasoning
+* Assumptions and uncertainty
+* Risks or tradeoffs
+* Recommended next step

collections/hve-core-all.collection.md

@@ -21,6 +21,9 @@ Supporting subagents included:
 
 - **Codebase Researcher** — Searches workspace for code patterns, conventions, and implementations
 - **External Researcher** — Retrieves external documentation, SDK references, and code samples
+- **GPT-5.4 Councilor** — Provides one independent frontier-model answer for council synthesis
+- **Opus 4.6 Councilor** — Provides one independent frontier-model answer for council synthesis
+- **Gemini 3.1 Pro Councilor** — Provides one independent frontier-model answer for council synthesis
 - **Phase Implementor** — Executes single implementation phases with change tracking
 - **Artifact Validator** — Validates implementation work against plans and conventions
 - **Prompt Tester** — Tests prompt files by following them literally in a sandbox

collections/hve-core-all.collection.yml

@@ -40,6 +40,8 @@ items:
   kind: agent
 - path: .github/agents/github/github-backlog-manager.agent.md
   kind: agent
+- path: .github/agents/hve-core/council.agent.md
+  kind: agent
 - path: .github/agents/hve-core/doc-ops.agent.md
   kind: agent
 - path: .github/agents/hve-core/memory.agent.md
@@ -50,8 +52,14 @@ items:
   kind: agent
 - path: .github/agents/hve-core/rpi-agent.agent.md
   kind: agent
+- path: .github/agents/hve-core/subagents/gemini-3-1-pro-councilor.agent.md
+  kind: agent
+- path: .github/agents/hve-core/subagents/gpt-5-4-councilor.agent.md
+  kind: agent
 - path: .github/agents/hve-core/subagents/implementation-validator.agent.md
   kind: agent
+- path: .github/agents/hve-core/subagents/opus-4-6-councilor.agent.md
+  kind: agent
 - path: .github/agents/hve-core/subagents/phase-implementor.agent.md
   kind: agent
 - path: .github/agents/hve-core/subagents/plan-validator.agent.md

collections/hve-core.collection.md

@@ -3,6 +3,7 @@ HVE Core provides the flagship RPI (Research, Plan, Implement, Review) workflow
 This collection includes agents for:
 
 - **RPI Agent** — Autonomous orchestrator that drives the full four-phase workflow
+- **LLM Council** — Dispatches the same question to GPT-5.4, Opus 4.6, and Gemini 3.1 Pro, then synthesizes the result
 - **Task Researcher** — Gathers context, discovers patterns, and produces research documents
 - **Task Planner** — Creates detailed implementation plans from research findings
 - **Task Implementor** — Executes plans with progressive tracking and change records
@@ -20,6 +21,9 @@ Supporting subagents included:
 
 - **Codebase Researcher** — Searches workspace for code patterns, conventions, and implementations
 - **External Researcher** — Retrieves external documentation, SDK references, and code samples
+- **GPT-5.4 Councilor** — Provides one independent frontier-model answer for council synthesis
+- **Opus 4.6 Councilor** — Provides one independent frontier-model answer for council synthesis
+- **Gemini 3.1 Pro Councilor** — Provides one independent frontier-model answer for council synthesis
 - **Phase Implementor** — Executes single implementation phases with change tracking
 - **Artifact Validator** — Validates implementation work against plans and conventions
 - **Prompt Tester** — Tests prompt files by following them literally in a sandbox

collections/hve-core.collection.yml

@@ -23,6 +23,8 @@ items:
     kind: agent
   - path: .github/agents/hve-core/doc-ops.agent.md
     kind: agent
+  - path: .github/agents/hve-core/council.agent.md
+    kind: agent
   - path: .github/agents/hve-core/prompt-builder.agent.md
     kind: agent
   - path: .github/agents/hve-core/task-researcher.agent.md
@@ -43,6 +45,12 @@ items:
     kind: agent
   - path: .github/agents/hve-core/subagents/phase-implementor.agent.md
     kind: agent
+  - path: .github/agents/hve-core/subagents/gpt-5-4-councilor.agent.md
+    kind: agent
+  - path: .github/agents/hve-core/subagents/opus-4-6-councilor.agent.md
+    kind: agent
+  - path: .github/agents/hve-core/subagents/gemini-3-1-pro-councilor.agent.md
+    kind: agent
   - path: .github/agents/hve-core/subagents/prompt-evaluator.agent.md
     kind: agent
   - path: .github/agents/hve-core/subagents/prompt-tester.agent.md

plugins/hve-core-all/README.md

@@ -28,6 +28,9 @@ Supporting subagents included:
 
 - **Codebase Researcher** — Searches workspace for code patterns, conventions, and implementations
 - **External Researcher** — Retrieves external documentation, SDK references, and code samples
+- **GPT-5.4 Councilor** — Provides one independent frontier-model answer for council synthesis
+- **Opus 4.6 Councilor** — Provides one independent frontier-model answer for council synthesis
+- **Gemini 3.1 Pro Councilor** — Provides one independent frontier-model answer for council synthesis
 - **Phase Implementor** — Executes single implementation phases with change tracking
 - **Artifact Validator** — Validates implementation work against plans and conventions
 - **Prompt Tester** — Tests prompt files by following them literally in a sandbox
@@ -66,12 +69,16 @@ copilot plugin install hve-core-all@hve-core
 | pptx                         | Creates, updates, and manages PowerPoint slide decks using YAML-driven content with python-pptx                                                                                                                                                                                                    |
 | pptx-subagent                | Executes PowerPoint skill operations including content extraction, YAML creation, deck building, and visual validation                                                                                                                                                                             |
 | github-backlog-manager       | Orchestrator agent for GitHub backlog management workflows including triage, discovery, sprint planning, and execution - Brought to you by microsoft/hve-core                                                                                                                                      |
+| council                      | LLM council that asks GPT-5.4, Opus 4.6, and Gemini 3.1 Pro the same question, then synthesizes a recommendation - Brought to you by microsoft/hve-core                                                                                                                                            |
 | doc-ops                      | Autonomous documentation operations agent for pattern compliance, accuracy verification, and gap detection - Brought to you by microsoft/hve-core                                                                                                                                                  |
 | memory                       | Conversation memory persistence for session continuity - Brought to you by microsoft/hve-core                                                                                                                                                                                                      |
 | pr-review                    | Comprehensive Pull Request review assistant ensuring code quality, security, and convention compliance - Brought to you by microsoft/hve-core                                                                                                                                                      |
 | prompt-builder               | Prompt engineering assistant with phase-based workflow for creating and validating prompts, agents, and instructions files - Brought to you by microsoft/hve-core                                                                                                                                  |
 | rpi-agent                    | Autonomous RPI orchestrator running Research → Plan → Implement → Review → Discover phases, using specialized subagents when task difficulty warrants them - Brought to you by microsoft/hve-core                                                                                                  |
+| gemini-3-1-pro-councilor     | Subagent councilor that answers a shared question independently using Gemini 3.1 Pro for the LLM Council agent - Brought to you by microsoft/hve-core                                                                                                                                              |
+| gpt-5-4-councilor            | Subagent councilor that answers a shared question independently using GPT-5.4 for the LLM Council agent - Brought to you by microsoft/hve-core                                                                                                                                                     |
 | implementation-validator     | Validates implementation quality against architectural requirements, design principles, and code standards with severity-graded findings - Brought to you by microsoft/hve-core                                                                                                                    |
+| opus-4-6-councilor           | Subagent councilor that answers a shared question independently using Opus 4.6 for the LLM Council agent - Brought to you by microsoft/hve-core                                                                                                                                                    |
 | phase-implementor            | Executes a single implementation phase from a plan with full codebase access and change tracking - Brought to you by microsoft/hve-core                                                                                                                                                            |
 | plan-validator               | Validates implementation plans against research documents, updating the Planning Log Discrepancy Log section with severity-graded findings - Brought to you by microsoft/hve-core                                                                                                                  |
 | prompt-evaluator             | Evaluates prompt execution results against Prompt Quality Criteria with severity-graded findings and categorized remediation guidance                                                                                                                                                              |
@@ -252,6 +259,7 @@ copilot plugin install hve-core-all@hve-core
 | markdown.instructions                             | Required instructions for creating or editing any Markdown (.md) files                                                                                                                                                                                      |
 | prompt-builder.instructions                       | Authoring standards for prompt engineering artifacts including prompts, agents, instructions, and skills                                                                                                                                                    |
 | pull-request.instructions                         | Required instructions for pull request description generation and optional PR creation using diff analysis, subagent review, and MCP tools - Brought to you by microsoft/hve-core                                                                           |
+| validation.instructions                           | Validation guidance for code, skill, and workflow edits                                                                                                                                                                                                     |
 | writing-style.instructions                        | Required writing style conventions for voice, tone, and language in all markdown content                                                                                                                                                                    |
 | jira-backlog-discovery.instructions               | Discovery protocol for Jira backlog management with user-centric, artifact-driven, and JQL-based issue discovery - Brought to you by microsoft/hve-core                                                                                                     |
 | jira-backlog-planning.instructions                | Reference specification for Jira backlog management tooling, planning files, search conventions, similarity assessment, and state persistence - Brought to you by microsoft/hve-core                                                                        |

plugins/hve-core-all/agents/hve-core/council.md

@@ -0,0 +1 @@
+../../../../.github/agents/hve-core/council.agent.md

plugins/hve-core-all/agents/hve-core/subagents/gemini-3-1-pro-councilor.md

@@ -0,0 +1 @@
+../../../../../.github/agents/hve-core/subagents/gemini-3-1-pro-councilor.agent.md

plugins/hve-core-all/agents/hve-core/subagents/gpt-5-4-councilor.md

@@ -0,0 +1 @@
+../../../../../.github/agents/hve-core/subagents/gpt-5-4-councilor.agent.md

plugins/hve-core-all/agents/hve-core/subagents/opus-4-6-councilor.md

@@ -0,0 +1 @@
+../../../../../.github/agents/hve-core/subagents/opus-4-6-councilor.agent.md