Skip to content
Aphelion

Agents Reference: Discovery Domain

Language: English | 日本語 Last updated: 2026-04-26 Update history:

  • 2026-04-26: Sync with #74 (issue #77)
  • 2026-04-25: split from Agents-Reference.md; #42 Audience: Agent developers

This page is one of five pages split from the original Agents-Reference.md (#42). It covers the Discovery domain agents. See the sibling pages for other domains: Orchestrators & Cross-Cutting, Delivery, Operations, Maintenance.

Table of Contents

Section titled “Table of Contents”

Discovery Domain

Section titled “Discovery Domain”

The Discovery domain (6 agents) handles requirements exploration.

interviewer

Section titled “interviewer”
  • Canonical: .claude/agents/interviewer.md
  • Domain: Discovery
  • Responsibility: Conducts requirements interviews, discovers implicit requirements, determines PRODUCT_TYPE and HAS_UI, generates INTERVIEW_RESULT.md. Runs on all Discovery plans.
  • Inputs: User’s project description (from discovery-flow)
  • Outputs: INTERVIEW_RESULT.md
  • AGENT_RESULT fields: PRODUCT_TYPE, HAS_UI, REQUIREMENTS_COUNT, IMPLICIT_REQUIREMENTS
  • NEXT conditions:
    • Minimal plan → done
    • Light plan → scope-planner (skips scope-planner in light — actually rules-designer)
    • Standard / Full plan → researcher
    • Rollback from poc-engineer → re-interview infeasible requirements

researcher

Section titled “researcher”
  • Canonical: .claude/agents/researcher.md
  • Domain: Discovery
  • Responsibility: Researches domain knowledge, competitors, external APIs, and technical risks. Defines ubiquitous language. Runs on Standard and Full plans.
  • Inputs: INTERVIEW_RESULT.md
  • Outputs: RESEARCH_RESULT.md
  • AGENT_RESULT fields: RISKS_FOUND, EXTERNAL_DEPS, COMPETITORS_ANALYZED, UBIQUITOUS_TERMS
  • NEXT conditions:
    • Standard / Full plan → poc-engineer
    • Rolled back from scope-planner → scope-planner

poc-engineer

Section titled “poc-engineer”
  • Canonical: .claude/agents/poc-engineer.md
  • Domain: Discovery
  • Responsibility: Validates technical feasibility with minimal PoC code. Identifies infeasible requirements and proposes alternatives. Runs on Standard and Full plans.
  • Inputs: INTERVIEW_RESULT.md, RESEARCH_RESULT.md
  • Outputs: POC_RESULT.md, code under poc/ directory
  • AGENT_RESULT fields: VERIFIED, BLOCKED_ITEMS, TECH_RECOMMENDATION
  • NEXT conditions:
    • No infeasible requirements, Full plan + HAS_UI → concept-validator
    • No infeasible requirements, otherwise → rules-designer
    • Infeasible requirements found → interviewer (rollback, STATUS: blocked)

concept-validator

Section titled “concept-validator”
  • Canonical: .claude/agents/concept-validator.md
  • Domain: Discovery
  • Responsibility: Validates UI/UX concept through wireframes and user flow diagrams. Runs only on Full plan when HAS_UI: true.
  • Inputs: INTERVIEW_RESULT.md, RESEARCH_RESULT.md (optional), POC_RESULT.md (optional)
  • Outputs: CONCEPT_VALIDATION.md
  • AGENT_RESULT fields: SCREENS, UX_ISSUES, IMPROVEMENTS
  • NEXT conditions: scope-planner

rules-designer

Section titled “rules-designer”
  • Canonical: .claude/agents/rules-designer.md
  • Domain: Discovery
  • Responsibility: Interactively determines project-specific coding conventions, Git workflow, build commands, and Repository declaration (Remote type: github | gitlab | gitea | local-only | none, consumed by git-rules.md Startup Probe). Generates .claude/rules/project-rules.md. Runs on Light and above.
  • Inputs: INTERVIEW_RESULT.md, RESEARCH_RESULT.md (optional), POC_RESULT.md (optional)
  • Outputs: .claude/rules/project-rules.md
  • AGENT_RESULT fields: LANGUAGE, FRAMEWORK, COMMIT_STYLE, BRANCH_STRATEGY, REPO_REMOTE_TYPE
  • NEXT conditions: scope-planner

scope-planner

Section titled “scope-planner”
  • Canonical: .claude/agents/scope-planner.md
  • Domain: Discovery
  • Responsibility: Defines MVP, prioritizes requirements with MoSCoW, assesses risks and costs, determines handoff readiness, and generates DISCOVERY_RESULT.md. Runs on Light and above.
  • Inputs: INTERVIEW_RESULT.md, RESEARCH_RESULT.md, POC_RESULT.md, CONCEPT_VALIDATION.md (as available)
  • Outputs: SCOPE_PLAN.md, DISCOVERY_RESULT.md
  • AGENT_RESULT fields: MVP_SCOPE, MUST_COUNT, SHOULD_COUNT, RISKS, HANDOFF_READY
  • NEXT conditions:
    • HANDOFF_READY: true → done
    • HANDOFF_READY: false → researcher (rollback, STATUS: blocked)
Section titled “Related Pages”

Canonical Sources

Section titled “Canonical Sources”