HirokiYasui

Use this agent for Phase 3 (Design) of the claude-forge. Synthesizes analysis and investigation findings into a concrete design document covering approach, architecture changes, data model, test strategy, and risk mitigation.

Use this agent for Phase 7 (Comprehensive Review) of the claude-forge. Reviews ALL implemented tasks holistically for cross-cutting concerns — naming consistency, code duplication, interface coherence, error handling patterns, and test coverage gaps. Fixes issues directly and reports findings.

Use this agent for Phase 3b (Design AI Review) of the claude-forge. Critically reviews a design document for spec alignment, coverage, completeness, consistency, test strategy, contradictions, and scope creep before human review. Outputs APPROVE, APPROVE_WITH_NOTES, or REVISE.

Use this agent for Phase 5 (Implementation) of the claude-forge. Implements a single task from the task list using TDD methodology — writes tests first, then implementation, then verifies. Writes an implementation summary upon completion.

Use this agent for Phase 6 (Implementation Review) of the claude-forge. Reviews a completed task's implementation against acceptance criteria, design alignment, test quality, code quality, and regression status. Outputs PASS, PASS_WITH_NOTES, or FAIL.

Use this agent for Phase 2 (Investigation) of the claude-forge. Goes deeper than situation analysis to examine root causes, edge cases, integration points, prior art, and deletion/rename impacts. Returns findings and open questions.

Use this agent for Phase 1 (combined Situation Analysis + Investigation) of the claude-forge when effort is S. Performs read-only codebase exploration covering both current-state mapping and deep-dive investigation in a single pass. Returns a structured markdown report that the architect can work from directly.

Use this agent for Phase 1 (Situation Analysis) of the claude-forge. Performs read-only codebase exploration to map the current state of files, interfaces, types, data flows, and tests relevant to a given task. Returns a structured markdown index.

Use this agent for Phase 4 (Task Decomposition) of the claude-forge. Breaks a design document into a numbered, dependency-aware task list with acceptance criteria, parallel/sequential markers, and file assignments.

Use this agent for Phase 4b (Tasks AI Review) of the claude-forge. Critically reviews a task list for design coverage, missing deletions, test gaps, dependency correctness, parallel safety, and acceptance criteria quality. Outputs APPROVE, APPROVE_WITH_NOTES, or REVISE.

Use this agent for Final Verification in the claude-forge. Runs full typecheck and test suite on the feature branch and reports results. Fixes failures if found.

Configure per-repository default flags for /forge pipelines. Interactive setup that saves preferences to .specs/preferences.json.

Orchestrate a full development pipeline using MCP-driven subagents. Accepts plain text or issue tracker URLs (GitHub, Jira, Linear, etc.) as input.