Agent skill
dev-step-all
Execute all remaining steps from split plan files without user confirmation. Automatically loops through steps until completion or error.
Install this agent skill to your Project
npx add-skill https://github.com/uuta/dotfiles/tree/main/skills/dev-step-all
SKILL.md
Dev Step All
stepファイルを全て読み込み、ユーザー確認なしで連続実装する。エラー・テスト失敗時は停止して報告。
Workflow Schema
See workflow.yaml for YAML structure.
Procedure
1. Validation
-
Get branch name from git:
git branch --show-current- Extract number:
feat/#1473→1473
- Extract number:
-
Check if
docs/{branch_name}/dev-workflow.yamlexists- If not exists → Ask user:
「dev-workflow.yaml が見つかりません。 先に /dev-plan を実行して計画を作成しますか?」 - Stop and wait for user response
- If not exists → Ask user:
-
Read YAML and validate state:
- If
plan.status!=Done→「計画が未完了です(status: {current_status})。 先に /dev-plan を完了してください。」 - Stop
- If
-
Check
execute.status:- If
Done→ "全stepが完了しています。再実行しますか?" - If
Pending→ Update toIn progress, proceed - If
In progress→ Resume fromcurrent_step
- If
2. Load Step File
-
Determine step file path:
- Project: Extract from pwd (e.g.,
mu-muc-app-service2→mu-muc-app-service) - Path:
~/uuta/Projects/{project}/{branch_name}/*-step-{current_step:02d}.md
- Project: Extract from pwd (e.g.,
-
Read step file using Glob + Read
- If not found → Error with available files list
-
Display step info:
## Step {current}/{total}: {step_title} 実装中...
3. Execute Step
-
Use Claude Code Agent Teams (TeamCreate + Task tool) as much as possible to parallelize work within a step:
- Identify independent sub-tasks (e.g., writing tests for different modules, implementing unrelated files, running checks)
- Spawn teammates to handle sub-tasks concurrently
- Coordinate results before marking the step complete
- Fall back to sequential execution only when tasks have strict dependencies
-
Implement according to step file instructions
-
Follow TDD approach if applicable:
- Red: Write failing tests first
- Green: Implement minimum code to pass
- Refactor: Clean up
-
Run verification checks defined in step file
- On test failure or error → STOP immediately and report the issue. Do not continue to the next step.
4. Complete Step
-
Update step file status to
Done(in Obsidian) -
Update YAML:
yamlexecute: current_step: {current + 1} -
Check if all steps complete:
- If
current_step > total_steps:yamlReport: "全てのstepが完了しました!" → STOPexecute: status: Done
- If
5. Auto-Continue
- If steps remain → Loop back to Phase 2 immediately (no user confirmation)
- Only stop when:
- All steps are
Done - An error or test failure occurs (report the issue)
- All steps are
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
pbi-sub-issue
タスク分割プラン(pbi-task-splitで作成)を元に、GitHub の Sub-issue を作成する。親 Issue との関係も自動で設定。
dev
Orchestrate TDD-based development workflow. Checks workflow state and delegates work to subagents (dev-plan, dev-step, create-pr).
neo-frontend-design
Create dark command-center / cyberpunk dashboard interfaces inspired by the OJPP Portal design. Use this skill when the user asks for a dark, neon-accented, terminal-aesthetic UI. Generates production-grade code with glitch effects, monospace typography, and per-module neon color coding.
tdd-plan
Create implementation plans following TDD methodology with test-first approach. Each test file is immediately followed by its implementation (fine-grained RED→GREEN cycles).
prototype
Generate 3 HTML prototype variants using a team of parallel agents. Each agent creates a distinct design pattern (e.g., hover effects, animations, layouts) based on user prompts or docs/goal.md. Use when the user wants to explore multiple design directions for a UI component.
review-format
Output in the specified format when a review of specific markdown is requested.
Didn't find tool you were looking for?