plan-sprint
について
このClaudeスキルは、開発者がアジャイルスプリントを計画する際に、バックログの洗練、ゴールの定義、チームキャパシティの計算、選択項目のタスク分解を支援します。構造化された`SPRINT-PLAN.md`ファイルを自動生成し、ゴール、選択項目、タスク分解、キャパシティ配分を含めます。新規スプリントの開始、スコープ変更後の再計画、バックログ調整後のスプリントサイクル確立にご活用ください。
クイックインストール
Claude Code
推奨npx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/plan-sprintこのコマンドをClaude Codeにコピー&ペーストしてスキルをインストールします
ドキュメント
Plan a Sprint
Plan time-boxed sprint: pick refined backlog items up to capacity, define clear goal, decompose into actionable tasks. Output: complete plan guiding team for sprint duration.
Use When
- New sprint in Scrum/agile project
- Re-plan after major scope change
- Ad-hoc → structured sprint cadence
- Post-backlog-grooming when items ready
- First sprint after charter approval
In
- Required: Product backlog (prioritized, w/ estimates)
- Required: Sprint duration (typically 1-2 wks)
- Required: Team members + availability
- Optional: Prior sprint velocity (story points or items completed)
- Optional: Sprint number + date range
- Optional: Carry-over from prev sprint
Do
Step 1: Review + Refine Backlog Items
Read current BACKLOG.md. For each candidate near top, verify:
- Clear title + desc
- Acceptance criteria (testable)
- Estimate (story points or T-shirt)
- No unresolved blockers
Refine missing. Split items > half sprint capacity → smaller pieces.
→ Top 10-15 items "sprint-ready" w/ acceptance criteria + estimates.
If err: items lack acceptance → write now. Can't estimate → schedule refinement, only pick ready.
Step 2: Define Sprint Goal
One sentence stating sprint outcome. Goal should:
- Achievable in sprint duration
- Valuable to stakeholders
- Testable (verifiable at sprint end)
**Sprint Goal**: [One sentence describing the objective]
Example: "Enable users to reset their password through email verification with two-factor authentication."
→ Sprint goal = one clear testable sentence.
If err: no coherent goal → backlog priorities scattered, consult product owner → focus on single valuable outcome.
Step 3: Calc Team Capacity
Available person-days per member:
## Team Capacity
| Team Member | Available Days | Overhead (%) | Net Capacity |
|-------------|---------------|-------------|--------------|
| [Name] | [Sprint days - PTO] | 20% | [Available × 0.8] |
| [Name] | [Sprint days - PTO] | 20% | [Available × 0.8] |
| **Total** | | | **[Sum] person-days** |
Overhead = meetings, reviews, ad-hoc (typically 15-25%).
Story points → use prior velocity. First sprint → 60-70% theoretical max.
→ Capacity calc'd in person-days or story points w/ doc'd assumptions.
If err: no historical velocity → conservative: 60%, adjust after. Better under-commit + deliver than over-commit + fail.
Step 4: Select Items + Compose Sprint Backlog
Pick from top of product backlog until capacity. Decompose each → tasks (2-8 hrs):
# Sprint Plan: Sprint [N]
## Document ID: SP-[PROJECT]-S[NNN]
### Sprint Details
- **Sprint Goal**: [From Step 2]
- **Duration**: [Start date] to [End date]
- **Capacity**: [From Step 3] person-days / [N] story points
- **Team**: [List team members]
### Sprint Backlog
| ID | Item | Points | Tasks | Assignee | Status |
|----|------|--------|-------|----------|--------|
| B-001 | [Item title] | 5 | 4 | [Name] | To Do |
| B-002 | [Item title] | 3 | 3 | [Name] | To Do |
| B-003 | [Item title] | 8 | 6 | [Name] | To Do |
| **Total** | | **16** | **13** | | |
### Task Breakdown
#### B-001: [Item title]
**Acceptance Criteria**: [From backlog item]
- [ ] Task 1: [Description] (4h, [Assignee])
- [ ] Task 2: [Description] (2h, [Assignee])
- [ ] Task 3: [Description] (4h, [Assignee])
- [ ] Task 4: [Description] (2h, [Assignee])
#### B-002: [Item title]
**Acceptance Criteria**: [From backlog item]
- [ ] Task 1: [Description] (3h, [Assignee])
- [ ] Task 2: [Description] (4h, [Assignee])
- [ ] Task 3: [Description] (2h, [Assignee])
#### B-003: [Item title]
**Acceptance Criteria**: [From backlog item]
- [ ] Task 1: [Description] (3h, [Assignee])
- [ ] Task 2: [Description] (4h, [Assignee])
- [ ] Task 3: [Description] (2h, [Assignee])
- [ ] Task 4: [Description] (3h, [Assignee])
- [ ] Task 5: [Description] (4h, [Assignee])
- [ ] Task 6: [Description] (2h, [Assignee])
### Risks and Dependencies
| Risk | Impact | Mitigation |
|------|--------|-----------|
| [Risk 1] | [Impact] | [Mitigation] |
| [Risk 2] | [Impact] | [Mitigation] |
### Carry-Over from Previous Sprint
| ID | Item | Reason | Remaining Effort |
|----|------|--------|-----------------|
| B-XXX | [Item] | [Reason] | [Hours/points] |
→ Sprint backlog w/ items up to capacity, each decomposed into tasks w/ time estimates.
If err: total points > capacity → drop lowest-pri item. Never exceed capacity by >10%. Deps block sequencing → reorder or defer.
Step 5: Document Commitments + Save
Write plan → SPRINT-PLAN.md (or SPRINT-PLAN-S[NNN].md for archive). Confirm:
- Sprint goal achievable w/ selected items
- No member overallocated (>100% capacity)
- Deps sequenced correctly
- Carry-over in capacity
- All acceptance criteria copied from backlog
Final validation:
# Check that total task hours align with capacity
grep -A 100 "Task Breakdown" SPRINT-PLAN.md | grep -o '([0-9]*h' | sed 's/[^0-9]//g' | awk '{sum+=$1} END {print "Total hours:", sum}'
→ SPRINT-PLAN.md created w/ complete backlog + task breakdown. Total hours ≤80% of available person-days × 8 hrs.
If err: commitments don't align w/ goal → revisit Step 4. Task hours > capacity → drop last item or decompose more granular.
Check
- Sprint goal = one clear testable sentence
- Capacity calc'd w/ doc'd assumptions (overhead %, PTO)
- Selected items don't exceed capacity (points or person-days)
- Every item has acceptance criteria in task breakdown
- Every item decomposed → tasks (2-8 hrs each)
- No member overallocated >100% capacity
- Carry-over doc'd w/ remaining effort
- Deps sequenced correctly
- Risks + mitigations doc'd
- SPRINT-PLAN.md created + saved
Traps
- No sprint goal: No goal → just bag of tasks. Goal = focus + basis for mid-sprint scope decisions.
- Over-commit: 100% capacity ignores interrupts, bugs, overhead. Plan 70-80% → buffer for unexpected.
- Tasks too large: >8 hrs hides complexity, hard tracking. Decompose to 2-8 hrs.
- Ignore carry-over: Unfinished items consume current sprint capacity. Account explicitly.
- Goal as item list: "Complete B-001, B-002, B-003" ≠ goal. Goal = outcome: "Users can reset password via email verification."
- No task ownership: Every task → assignee at planning → surface capacity conflicts early.
- Skip acceptance criteria: Tasks w/o criteria = untestable. Copy criteria from backlog into task breakdown.
→
manage-backlog— maintain + prioritize backlog feeding planningdraft-project-charter— project context + initial scope for first sprintgenerate-status-report— report progress + velocity to stakeholdersconduct-retrospective— review sprint, improve planningcreate-work-breakdown-structure— WBS work packages feed backlog in hybrid agile-waterfall
GitHub リポジトリ
関連スキル
llamaguard
その他LlamaGuardは、暴力やヘイトスピーチなど6つの安全性カテゴリーにおいて、LLMの入力と出力をモデレートするMetaの70-80億パラメータモデルです。94〜95%の精度を提供し、vLLM、Hugging Face、Amazon SageMakerを使用してデプロイ可能です。このスキルを使用して、AIアプリケーションにコンテンツフィルタリングと安全策を簡単に統合できます。
cost-optimization
その他このClaudeスキルは、リソースの適正サイジング、タグ付け戦略、支出分析を通じて、開発者がクラウドコストを最適化することを支援します。AWS、Azure、GCPにわたるクラウド支出の削減とコストガバナンスの実施のためのフレームワークを提供します。インフラコストの分析、リソースの適正サイジング、または予算制約への対応が必要な際にご利用ください。
quantizing-models-bitsandbytes
その他このスキルは、bitsandbytesを使用してLLMを8ビットまたは4ビット精度に量子化し、精度の低下を最小限に抑えつつ50〜75%のメモリ削減を実現します。限られたGPUメモリでより大規模なモデルを実行したり、推論を高速化するのに理想的で、INT8、NF4、FP4などのフォーマットをサポートしています。HuggingFace Transformersと統合され、QLoRAトレーニングや8ビットオプティマイザーを可能にします。
dispatching-parallel-agents
その他このClaudeスキルは、複数のエージェントを配備し、3つ以上の独立した問題を並行して調査・修正します。共有状態や依存関係がなく解決可能な、無関係な障害が発生するシナリオ向けに設計されています。中核となる機能は並列問題解決であり、効率を最大化するために独立した問題領域ごとに1つのエージェントを割り当てます。
