name	dev-medium
description	Multi-agent problem solving with architect and planner subagents for medium complexity tasks.
allowed-tools	Read, Edit, Write, Glob, Grep, Bash, Agent, mcp__sequential-thinking__sequentialthinking

Use sequential-thinking mcp and all its tools that you will need about the problem and how to solve it. You must ultrathink for the solution and use reasoning.

You must consider edge cases and follow best coding practices for everything. Never do bandaid fixes.

Configuration

STEP 1: Launch the architect agent in parallel with the following prompt:

Investigate the problem and the relevant codebase.
Analyze the current architecture, data flow, and affected components.
Write an "INVESTIGATION_REPORT.md" inside the claude-code-storage/claude-instance-{id}/ directory covering: current state analysis, affected files, root cause or feature scope, and integration points.

STEP 2: Once the architect agent completes, launch the planner agent with the following prompt:

Read the "INVESTIGATION_REPORT.md" from the claude-code-storage/claude-instance-{id}/ directory.
Based on the investigation, create a "PLAN.md" in the same directory covering: implementation steps with file paths, dependency order, testing strategy, risks and mitigations.

STEP 3: After the planner agent finishes, enter plan mode and read the "PLAN.md" file and present the plan to the user so that they can either accept or adjust it.

Problem: $ARGUMENTS

同仓库更多 Skills

同仓库

sync-claude-ansible

terakoya76/terakoya76-playbooks

ローカル ~/.claude/ 配下の Claude Code 設定 (CLAUDE.md, settings.json, mcp.json, agents/, hooks/, rules/, skills/) を、ansible が管理する terakoya76-playbooks リポジトリの roles/dotfiles/files/claude/ にミラーする。「ansible に反映」「claude config を sync」「dotfiles に取り込み」等と依頼されたとき、あるいは local で skill / agent / settings を更新した直後に使う。

2026-05-200

commit

terakoya76/terakoya76-playbooks

Check diffs and commit with conventional commits guidelines. No co-authored or generated-by lines.

2026-05-200

dev-easy

terakoya76/terakoya76-playbooks

Use sequential-thinking MCP for problem solving and error fixing with ultrathink reasoning. Considers edge cases and best practices. Runs security review after code changes.

2026-05-200

explain-code

terakoya76/terakoya76-playbooks

Analyze and explain complex code sections in any programming language with clear, comprehensive explanations.

2026-05-200

refactor-clean

terakoya76/terakoya76-playbooks

Initiate refactor-cleaner subagent to safely identify and remove dead code with test verification.

2026-05-200

repro

terakoya76/terakoya76-playbooks

Initiate bug-reproducer subagent to create reproducing cases and narrow down to minimal reproduction.

2026-05-200

name	dev-medium
description	Multi-agent problem solving with architect and planner subagents for medium complexity tasks.
allowed-tools	Read, Edit, Write, Glob, Grep, Bash, Agent, mcp__sequential-thinking__sequentialthinking

Use sequential-thinking mcp and all its tools that you will need about the problem and how to solve it. You must ultrathink for the solution and use reasoning.

You must consider edge cases and follow best coding practices for everything. Never do bandaid fixes.

Configuration

STEP 1: Launch the architect agent in parallel with the following prompt:

Investigate the problem and the relevant codebase.
Analyze the current architecture, data flow, and affected components.
Write an "INVESTIGATION_REPORT.md" inside the claude-code-storage/claude-instance-{id}/ directory covering: current state analysis, affected files, root cause or feature scope, and integration points.

STEP 2: Once the architect agent completes, launch the planner agent with the following prompt:

Read the "INVESTIGATION_REPORT.md" from the claude-code-storage/claude-instance-{id}/ directory.
Based on the investigation, create a "PLAN.md" in the same directory covering: implementation steps with file paths, dependency order, testing strategy, risks and mitigations.

STEP 3: After the planner agent finishes, enter plan mode and read the "PLAN.md" file and present the plan to the user so that they can either accept or adjust it.

Problem: $ARGUMENTS