inquiry

name	inquiry
description	Orchestrate multi-agent deliberation for INQ work items through all 4 phases
aliases	["inq","deliberate"]

Skill: Inquiry Orchestration

You are the Inquiry Orchestration agent. Your goal is to guide INQ work items through all four deliberation phases—Research, Synthesis, Debate, and Consensus—coordinating multiple agents to reach well-reasoned decisions.

Overview

The /inquiry skill manages the complete lifecycle of an INQ (Inquiry) work item:

1. Phase 1: Research - Spawn independent agents to explore the problem space
2. Phase 2: Synthesis - Consolidate findings into unified understanding
3. Phase 3: Debate - Resolve conflicts through structured argumentation
4. Phase 4: Consensus - Formalize decisions and spawn FEAT work item(s)

Usage

# Start or continue an inquiry
/inquiry INQ-001

# Start a specific phase
/inquiry INQ-001 --phase research
/inquiry INQ-001 --phase synthesis
/inquiry INQ-001 --phase debate
/inquiry INQ-001 --phase consensus

# With agent configuration
/inquiry INQ-001 --agents 5 --model claude

# Fallback mode (no ccmux)
/inquiry INQ-001 --mode manual

Arguments

Argument	Required	Description
inquiry_id	Yes	The inquiry ID (e.g., INQ-001) or path to inquiry directory
--phase	No	Start/resume at specific phase (auto-detects if omitted)
--agents	No	Number of research agents (overrides inquiry_report.json)
--model	No	Model for spawned agents (claude, gemini, codex)
--mode	No	Execution mode: ccmux (default) or manual
--timeout	No	Timeout per phase in seconds (default: 600)

Workflow

1. Initialization

Read inquiry_report.json to determine:

• Current phase and status
• Number of research agents required
• Constraints and context

# Locate inquiry
INQUIRY_PATH=$(find_inquiry "$1")

# Load configuration
inquiry_report.json → phase, status, research_agents, constraints

2. Phase Detection

Automatically detect which phase to execute based on:

• Current phase field in inquiry_report.json
• Existence of phase artifacts (research/, SYNTHESIS.md, DEBATE.md, CONSENSUS.md)

if phase == "new" or phase == "research":
    → Execute Phase 1: Research
elif phase == "synthesis":
    → Execute Phase 2: Synthesis
elif phase == "debate":
    → Execute Phase 3: Debate
elif phase == "consensus":
    → Execute Phase 4: Consensus
elif phase == "completed":
    → Report completion, show spawned features

---

Phase 1: Research

Goal: Gather diverse perspectives through parallel, independent exploration.

With ccmux (Recommended)

1. Generate Prompts via /inquiry-prompts:

   /inquiry-prompts $INQUIRY_PATH --algorithm round-robin --output json
   

2. Spawn Research Agents:

   For each agent prompt:
     ccmux_create_session(
       name: "inq-{inquiry_id}-agent-{N}",
       command: "claude --dangerously-skip-permissions",
       tags: ["worker", "{inquiry_id}"]
     )
     ccmux_send_input(pane_id, prompt)
   

3. Monitor Completion via /inquiry-collect:

   /inquiry-collect $INQUIRY_ID --mode ccmux --timeout $TIMEOUT
   

4. Update Status:

   {
     "phase": "synthesis",
     "status": "synthesis",
     "phase_history": [
       {"phase": "research", "entered_date": "...", "notes": "Completed N/N agents"}
     ]
   }
   

Without ccmux (Manual Mode)

1. Generate Prompts to files:

   /inquiry-prompts $INQUIRY_PATH --algorithm round-robin --output files
   

2. Display Instructions:

   Research prompts generated in research/ directory.
   
   To complete Phase 1:
   1. Open each research/agent-N.md in separate Claude sessions
   2. Have each agent complete their research independently
   3. Save outputs back to research/agent-N.md
   4. Run: /inquiry $INQUIRY_ID --phase synthesis
   

3. Collect When Ready:

   /inquiry-collect $INQUIRY_ID --mode file
   

Research Output Structure

Each research/agent-N.md contains:

• Problem analysis
• Approaches explored
• Evidence gathered
• Key findings
• Recommendations

---

Phase 2: Synthesis

Goal: Consolidate all research findings into a unified understanding.

Process

1. Load Research Reports:

   Read all research/agent-*.md files
   Read SUMMARY.md (generated by inquiry-collector)
   

2. Generate Synthesis Prompt:

   # Synthesis Agent - Consolidate Research Findings
   
   ## Task
   You are the Synthesis Agent for **{inquiry_title}**.
   
   Review all research reports and create a comprehensive synthesis.
   
   ## Research Reports
   {all_agent_reports}
   
   ## Summary of Findings
   {summary_md_content}
   
   ## Instructions
   1. Identify common themes across all reports
   2. Note areas of agreement and strong consensus
   3. Highlight conflicts and divergent perspectives
   4. Extract key decision points requiring debate
   5. Create SYNTHESIS.md following the template below
   
   ## Output Template
   
   # Synthesis: {inquiry_title}
   
   ## Overview
   [Executive summary of consolidated findings]
   
   ## Common Themes
   [Themes that emerged across multiple agents]
   
   ## Areas of Agreement
   [Points where agents converged]
   
   ## Areas of Disagreement
   [Points requiring debate]
   
   ## Key Decision Points
   [Decisions that need to be made]
   
   ## Evidence Summary
   [Consolidated evidence from all agents]
   
   ## Questions for Debate
   [Specific questions to resolve in Phase 3]
   

3. Execute Synthesis:

   • ccmux mode: Spawn synthesis agent, collect SYNTHESIS.md
   • manual mode: Display prompt, await user completion

4. Update Status:

   {
     "phase": "debate",
     "status": "debate"
   }
   

---

Phase 3: Debate

Goal: Resolve conflicts through structured adversarial argumentation.

Process

1. Load Synthesis:

   Read SYNTHESIS.md
   Extract "Areas of Disagreement" and "Key Decision Points"
   

2. Generate Debate Structure:

   # Debate: {inquiry_title}
   
   ## Decision Points
   
   ### Decision Point 1: {topic}
   
   #### Position A: {stance}
   **Advocate**: Agent A
   
   **Arguments**:
   1. [Argument with evidence]
   2. [Argument with evidence]
   
   **Supporting Evidence**:
   - [Evidence from research]
   
   #### Position B: {stance}
   **Advocate**: Agent B
   
   **Arguments**:
   1. [Counter-argument with evidence]
   2. [Counter-argument with evidence]
   
   **Supporting Evidence**:
   - [Evidence from research]
   
   #### Rebuttal Round
   
   **Agent A Response**:
   [Response to Position B arguments]
   
   **Agent B Response**:
   [Response to Position A arguments]
   
   #### Resolution
   **Prevailing Position**: [A or B]
   **Rationale**: [Why this position was chosen]
   **Confidence**: [High/Medium/Low]
   

3. Execute Debate:

   • ccmux mode: Spawn advocate agents for each position
   • manual mode: Facilitate structured debate prompt

4. Generate DEBATE.md: Record all arguments, counter-arguments, and resolutions.

5. Update Status:

   {
     "phase": "consensus",
     "status": "consensus"
   }
   

---

Phase 4: Consensus

Goal: Formalize decisions and spawn implementation work.

Process

1. Load Debate Results:

   Read DEBATE.md
   Extract all resolutions and rationales
   

2. Generate Consensus Document:

   # Consensus: {inquiry_title}
   
   ## Executive Summary
   [Summary of the final decision]
   
   ## Decision Record
   
   ### Decision 1: {topic}
   **Resolution**: {chosen_approach}
   **Rationale**: {why_chosen}
   **Alternatives Rejected**:
   - {alternative}: {reason_rejected}
   
   ## Implementation Plan
   
   ### Recommended Approach
   [Detailed description of chosen approach]
   
   ### Success Criteria
   - [Criterion 1]
   - [Criterion 2]
   
   ### Risks and Mitigations
   | Risk | Mitigation |
   |------|------------|
   | ... | ... |
   
   ## Work Items
   
   ### FEAT-XXX: {feature_title}
   **Description**: {description}
   **Priority**: {priority}
   **Estimated Effort**: {effort}
   
   ## Approval
   
   **Consensus Reached**: {date}
   **Source Inquiry**: {inquiry_id}
   **Participants**: {agent_list}
   

3. Spawn FEAT Work Item(s):

   /work-item-creation FEAT \
     --title "{feature_title}" \
     --component "{component}" \
     --description "Implementation of consensus from {inquiry_id}" \
     --source-inquiry "{inquiry_id}"
   

4. Update Inquiry:

   {
     "phase": "completed",
     "status": "completed"
      
   
   

---

Integration Points

Skills Used

Skill	Phase	Purpose
/inquiry-prompts	Research	Generate agent prompts from QUESTION.md
/inquiry-collect	Research	Collect and consolidate agent outputs
/work-item-creation	Consensus	Create FEAT work items

ccmux Tools

Tool	Purpose
ccmux_create_session	Spawn agent sessions
ccmux_send_input	Send prompts to agents
ccmux_get_status	Check agent completion
ccmux_read_pane	Extract agent output
ccmux_set_tags	Tag sessions with inquiry ID
ccmux_kill_session	Cleanup after completion

File Dependencies

File	Phase	Description
inquiry_report.json	All	Metadata and state tracking
QUESTION.md	Research	Problem statement and context
research/agent-N.md	Research	Independent research reports
SUMMARY.md	Research→Synthesis	Collector output, synthesis input
SYNTHESIS.md	Synthesis→Debate	Consolidated findings
DEBATE.md	Debate→Consensus	Structured arguments
CONSENSUS.md	Consensus	Final decisions

---

Error Handling

Phase 1: Research

Error	Behavior
ccmux unavailable	Fall back to manual mode
Agent timeout	Mark partial, continue with available
Prompt generation fail	Report error, suggest QUESTION.md fix

Phase 2: Synthesis

Error	Behavior
Missing research files	Report which agents incomplete
SUMMARY.md missing	Generate from available reports

Phase 3: Debate

Error	Behavior
No disagreements found	Skip debate, proceed to consensus
Resolution unclear	Flag for human review

Phase 4: Consensus

Error	Behavior
FEAT creation fails	Report error, allow manual creation
Missing resolutions	Block until debate complete

---

Examples

Complete Inquiry Workflow

# Start new inquiry
/inquiry INQ-001

# Output:
# Starting INQ-001: Authentication Strategy Decision
# Phase: Research (1/4)
#
# Generating research prompts...
# Spawning 3 research agents...
#   Agent 1: Security Analysis (session: inq-001-agent-1)
#   Agent 2: Implementation Options (session: inq-001-agent-2)
#   Agent 3: Cost Analysis (session: inq-001-agent-3)
#
# Monitoring agent completion...
# Agent 1: Complete (2847 chars)
# Agent 2: Complete (3102 chars)
# Agent 3: Complete (2956 chars)
#
# Research phase complete. Proceeding to synthesis...
# Phase: Synthesis (2/4)
#
# [... synthesis completes ...]
#
# Phase: Debate (3/4)
#
# [... debate completes ...]
#
# Phase: Consensus (4/4)
#
# Generating consensus document...
# Creating FEAT-024: Implement OAuth2 Authentication
#
# INQ-001 completed successfully!
# Spawned features: FEAT-024

Resume Partial Inquiry

# Resume inquiry stuck in synthesis
/inquiry INQ-001

# Output:
# Resuming INQ-001: Authentication Strategy Decision
# Current phase: synthesis
#
# Research artifacts found:
#   research/agent-1.md ✓
#   research/agent-2.md ✓
#   research/agent-3.md ✓
#   SUMMARY.md ✓
#
# Proceeding with synthesis phase...

Manual Mode

/inquiry INQ-001 --mode manual

# Output:
# Starting INQ-001 in manual mode (no ccmux)
#
# Phase 1: Research
#
# Generated research prompts:
#   research/agent-1.md - Security Analysis prompt
#   research/agent-2.md - Implementation Options prompt
#   research/agent-3.md - Cost Analysis prompt
#
# Instructions:
# 1. Open each file in a separate Claude session
# 2. Have each agent complete their research independently
# 3. Save their outputs back to the respective files
# 4. Run: /inquiry INQ-001 to continue
#
# Waiting for research completion...

---

Configuration

inquiry_report.json

{
  "inquiry_id": "INQ-001",
  "title": "Authentication Strategy Decision",
  "component": "auth",
  "priority": "P1",
  "status": "new",
  "phase": "research",
  "research_agents": 3,
  "constraints": [
    "Must support SSO",
    "Must comply with SOC2",
    "Budget: $10k/year max"
  ],
  "context": "Evaluating authentication approaches for new mobile app..."
}

Agent Tags (ccmux mode)

Sessions are tagged for tracking:

["worker", "INQ-001", "research"]
["worker", "INQ-001", "synthesis"]
["worker", "INQ-001", "debate-advocate-a"]
["worker", "INQ-001", "debate-advocate-b"]

---

Rules

1. Phase Order is Mandatory: Research → Synthesis → Debate → Consensus
2. Independence in Research: Agents must not share findings until synthesis
3. All Constraints Must Be Satisfied: No proposal can violate stated constraints
4. Debate Requires Positions: At least two perspectives per decision point
5. Consensus Creates Work: At least one FEAT must be spawned
6. State is Persistent: inquiry_report.json tracks all progress
7. Fallback is Available: Manual mode works without ccmux

Scripts

Supporting scripts in scripts/:

• phase_manager.py - Phase detection and transition logic
• synthesis_generator.py - Create synthesis prompts
• debate_structurer.py - Structure debate format
• consensus_builder.py - Generate consensus documents