スキル一覧に戻る
NoNameItem

linking-design

by NoNameItem

0🍴 0📅 2026年1月25日
GitHubで見るManusで実行

SKILL.md


name: linking-design description: Use after completing brainstorming/design phase. Links design document to task, parses subtasks, shows preview for approval, implements smart merge with existing subtasks. Handles post-design workflow for beads tasks.

Flow: After Design

Overview

Core principle: Preview before creating.

This skill handles the post-design workflow: linking design documents, extracting subtasks, and creating task hierarchy. Always shows what will be created and asks for confirmation - even when structure is clear and user seems in a hurry.

Quick Reference

StepActionKey Point
1. Find TaskGet in_progress leaf taskAsk if multiple
2. Find DesignNewest in docs/plans/Heuristic search
3. ParseExtract subtasks from designMultiple formats
4. PreviewShow what will be createdFormat + dependencies
5. ConfirmGet explicit approvalEven under pressure
6. MergeCheck existing subtasksSkip duplicates
7. CreateOnly new subtasksAfter confirmation
8. LinkSave design pathTo description

Scope: This skill ONLY handles design linking and subtask creation. Does NOT create branches, commit files, or update statuses - those are separate workflows.

Workflow

Follow these steps in order. Do not skip steps.

1. Find In-Progress Leaf Task

bd list --status=in_progress

Filter for leaf tasks (tasks without children or where all children are closed).

If multiple found: Ask user which task this design is for. If none found: Suggest running flow:start first.

2. Find Newest Design Document

ls -t docs/plans/*.md | head -1

Look in docs/plans/ for newest markdown file (by modification time).

Heuristics for "design" vs other docs:

  • Recent (within last hour is good signal)
  • Contains sections like "Requirements", "Architecture", "Tasks", "Implementation"
  • User just mentioned finishing design

If multiple candidates: Show list, let user choose. If none found: Ask user for design file path.

3. Parse Subtasks from Design

Heuristically detect multiple formats:

  • Numbered lists: 1. Do X, 2. Do Y
  • Bullet lists: - Do X, * Do Y
  • Checkboxes: - [ ] Do X, - [x] Do Y
  • Headers: ## Task: Do X, ### Do X
  • Sections: Look for "Implementation", "Tasks", "Subtasks", "Steps"

Extract for each subtask:

  • Title (required)
  • Description (if multi-line)
  • Priority (inherit from parent if not specified)

If no subtasks found: Ask: "No clear subtasks found in design. Should I:

  1. Look in a specific section (which one?)
  2. Treat this as atomic task (no subtasks needed)
  3. Let you manually specify subtasks"

4. Show Preview (MANDATORY)

Before creating anything, display preview:

Found 5 subtasks in design docs/plans/2026-01-16-feature-x.md:

1. {parent-id}.1: Implement input component
   Description: Create React component for user input, add form validation, handle submission
   Priority: P3 (inherits from parent)

2. {parent-id}.2: Add data validation
   Description: Implement validation rules, add error handling, write validation tests
   Priority: P3

3. {parent-id}.3: Create storage layer
   Description: Database schema, CRUD operations, migration scripts
   Priority: P3

4. {parent-id}.4: Write integration tests
   Description: Test full flow end-to-end, mock external dependencies, cover edge cases
   Priority: P3

5. {parent-id}.5: Update documentation
   Description: API docs, user guide, architecture diagram
   Priority: P3

All subtasks will depend on parent task {parent-id}.
Design will be linked in task description as: Design: docs/plans/2026-01-16-feature-x.md

Preview must include:

  • Count of subtasks
  • Full task IDs
  • Titles and descriptions
  • Priority levels
  • Dependencies
  • Design link that will be added

5. Ask for Confirmation (MANDATORY)

Always ask explicitly:

"Should I create these {N} subtasks? (yes/no)"

Wait for user response.

If user says no: Ask what they want to adjust. If user says yes: Proceed to step 6.

Do NOT:

  • Assume yes because "structure is clear"
  • Skip because user said "asap"
  • Create immediately because "being helpful"

6. Check Existing Subtasks (Merge Logic)

bd show {parent-id}

Check if parent task already has children.

If existing subtasks found:

Checking existing subtasks...

Found 2 existing:
- {parent-id}.1: Implement input component (exists)
- {parent-id}.2: Add data validation (exists)

From design (5 total):
✓ Skip: Implement input component (already exists)
✓ Skip: Add data validation (already exists)
+ New: Create storage layer
+ New: Write integration tests
+ New: Update documentation

Should I create these 3 NEW subtasks? (yes/no)

Merge rules:

  • Exact match (title identical) → Skip
  • Close match (title similar, >80% match) → Ask user
  • New → Include in creation

7. Create Subtasks

Only after confirmation, create subtasks:

bd create --title="Subtask title" --description="..." --priority=P3 --parent={parent-id}

Create each new subtask in order.

Output progress:

Creating subtasks...
✓ Created {parent-id}.3: Create storage layer
✓ Created {parent-id}.4: Write integration tests
✓ Created {parent-id}.5: Update documentation

Created 3 new subtasks under {parent-id}.

Update parent task description with design link:

bd update {parent-id} --description="{existing-description}\n\nDesign: docs/plans/{design-filename}"

If design link already exists: Ask: "Task already has design link: {old-link}. Update to new design {new-link}? (yes/no)"

Scope Boundaries

This Skill DOES:

✅ Find in_progress leaf task ✅ Find newest design document ✅ Parse subtasks heuristically ✅ Show preview with details ✅ Ask for confirmation ✅ Check existing subtasks (merge) ✅ Create new subtasks only ✅ Link design path to description

This Skill Does NOT:

❌ Create git branches (use flow:start) ❌ Commit design document (separate workflow) ❌ Update task status (use flow:start or manual bd update) ❌ Run tests or builds ❌ Implement code ❌ Make coffee

One skill, one job. Stay in scope.

Red Flags - STOP

If you're thinking any of these, STOP and follow the workflow:

  • "User said asap"
  • "I'll complete the setup"
  • "Structure is clear from design"
  • "No need to show preview"
  • "Confirmation is overhead"
  • "I'll just do it"
  • "I'll create the branch too"
  • "Let me commit the design"

All of these mean: Follow the workflow. Show preview. Ask confirmation. Stay in scope.

Common Rationalizations

ExcuseReality
"User wants to start asap"Preview takes 5 seconds. Fixing wrong subtasks takes 5 minutes.
"I'll complete the setup"Scope creep. One skill, one job.
"Structure is clear from design"User might want to adjust. Always show preview.
"No existing subtasks"Check anyway. Merge logic prevents duplicates.
"Confirmation slows things down"Creating wrong things is slower than confirming right things.
"This is being helpful"Assuming isn't helpful - consulting is.
"I'll create the branch too"Out of scope. Use flow:start for branches.

Examples

❌ BAD: Skip preview and confirmation

User: "I finished design, run after-design please, need to implement asap"
Agent: [Reads design, creates 5 subtasks immediately]
       Created 5 subtasks under claude-tools-abc. Also created feature branch.

Problems:

  • No preview shown
  • No confirmation requested
  • Out of scope (branch creation)
  • Time pressure bypassed workflow

✅ GOOD: Always preview and confirm

User: "I finished design, run after-design please, need to implement asap"
Agent: Found 5 subtasks in docs/plans/2026-01-16-feature-x.md:

       1. claude-tools-abc.1: Implement input component
          Description: React component, form validation, submission handling
          Priority: P3

       2. claude-tools-abc.2: Add data validation
          ...

       [shows all 5 with details]

       Should I create these 5 subtasks? (yes/no)

User: yes
Agent: Creating subtasks...
       ✓ Created 5 subtasks
       ✓ Linked design to task description

       Ready to start implementation! Use flow:start to choose a subtask.

Correct because:

  • Showed full preview
  • Asked for confirmation despite "asap"
  • Stayed in scope (no branch)
  • Guided to next step

❌ BAD: Ignore existing subtasks

User: "after-design" [task already has 2 subtasks]
Agent: [Creates all 5 subtasks from design, including 2 duplicates]

Problem: No merge logic, created duplicates.

✅ GOOD: Smart merge

User: "after-design" [task already has 2 subtasks]
Agent: Checking existing subtasks...

       Found 2 existing:
       - claude-tools-abc.1: Implement input component (exists)
       - claude-tools-abc.2: Add data validation (exists)

       From design (5 total):
       ✓ Skip: Implement input component (exists)
       ✓ Skip: Add data validation (exists)
       + New: Create storage layer
       + New: Write integration tests
       + New: Update documentation

       Should I create these 3 NEW subtasks? (yes/no)

Correct because:

  • Checked existing
  • Showed merge logic
  • Only proposes new subtasks
  • Still asks confirmation

Edge Cases

Multiple In-Progress Tasks

Found 2 in_progress leaf tasks:
1. claude-tools-abc: Feature X
2. claude-tools-def: Feature Y

Which task is this design for? (enter 1 or 2, or task ID)

No Subtasks in Design

No clear subtasks found in docs/plans/2026-01-16-design.md.

Should I:
1. Look in a specific section (which one?)
2. Treat this as atomic task (no subtasks needed)
3. Let you manually specify subtasks

What would you like to do?
Task already has design link: docs/plans/old-design.md

Update to new design docs/plans/2026-01-16-new-design.md? (yes/no)

The Bottom Line

Always follow the workflow. Preview before creating. Confirmation is not overhead - it's the service.

Parse heuristically, show clearly, ask explicitly, then act.

Stay in scope. One skill, one job, done well.

スコア

総合スコア

50/100

リポジトリの品質指標に基づく評価

SKILL.md

SKILL.mdファイルが含まれている

+20
LICENSE

ライセンスが設定されている

0/10
説明文

100文字以上の説明がある

0/10
人気

GitHub Stars 100以上

0/15
最近の活動

3ヶ月以内に更新がある

0/10
フォーク

10回以上フォークされている

0/5
Issue管理

オープンIssueが50未満

+5
言語

プログラミング言語が設定されている

+5
タグ

1つ以上のタグが設定されている

0/5

レビュー

💬

レビュー機能は近日公開予定です