Scrum Skill

Scrum is the MEANS, not the GOAL. The goal is delivering value to users.

Arguments: $ARGUMENTS

Version Check (on every invocation)

Before routing arguments, check for version mismatch:

1. If docs/scrum/ exists (not first-time setup):
   • Read metadata.version from this SKILL.md (currently: 1.3.2)
   • Read docs/scrum/.scrum-version (if exists)
   • If versions differ or .scrum-version is missing: display "スキル v{new} が利用可能です（現在 v{old}）。/scrum update で更新してください。" (If .scrum-version is missing, treat as unknown.)
   • Continue with the requested action regardless (version check is informational only)
2. If docs/scrum/ does not exist: proceed to setup (no version check needed)

Argument Routing

Detect first time: Check if docs/scrum/ directory exists in project root.

Install / Uninstall

/scrum install

Install this skill globally so /scrum works in any project:

1. Detect skill location: Find the directory containing this SKILL.md
2. Check existing installation: If ~/.claude/skills/scrum already exists, report "Already installed" and show the current link target
3. Create symlink: ln -s {SKILL_DIR} ~/.claude/skills/scrum
4. Verify: Confirm the link was created successfully

Example:

ln -s /path/to/scrum_agents/skills/scrum ~/.claude/skills/scrum

If the link target differs from the current skill directory, ask whether to update it.

/scrum uninstall

Remove the global skill symlink and optionally clean project files:

1. Remove symlink: rm ~/.claude/skills/scrum
2. Ask about project files: "プロジェクトのScrumファイルも削除しますか？"
   • If yes, remove:
     • docs/scrum/ directory
     • .claude/agents/scrum-*.md
     • .claude/rules/scrum-*.md
     • Scrum section in CLAUDE.md
   • If no, keep project files (they still work as documentation)
3. Report: Show what was removed

Tool Agnosticism

This skill defines Scrum process, NOT specific tools.

Scrum requires certain capabilities (backlog management, work tracking, increment review). HOW those capabilities are fulfilled depends on the project's environment.

Scrum Capabilities → Environment Mapping

Environment Detection (at setup)

During setup, detect what's available:

1. VCS: git? Other?
2. Remote platform: Check for existing skills, MCPs, or CLIs (e.g., gh, glab, bb)
3. Issue tracker: Integrated with platform? Separate (Jira, Redmine)?
4. User preference: Ask stakeholder if they have a preferred workflow

Adaptation Rules

• Always: Use docs/scrum/ for local Scrum records (sprint archives, logs, DoD). This is the source of truth for the Scrum process itself.
• If external tools are available: Use them for backlog management, work tracking, and async communication. Map Scrum concepts to the platform's native features.
• If no external tools: Use docs/scrum/backlog.md and docs/scrum/sprints/current.md as the full workflow. Everything works.
• If stakeholder specifies tools: Adapt to their environment. Ask for guidance on how to use unfamiliar tools, or recommend they install appropriate skills/MCPs.

Recommending Tools (not requiring them)

During setup, if no external tool is detected, suggest (not require):

"バックログ管理や非同期フィードバックのために、プロジェクト管理ツールの導入を検討できます。 例: GitHub Issues, GitLab Issues, Jira, Redmine など。 対応するスキルや MCP があれば導入をお勧めします。 なくても、マークダウンファイルで全て運用できます。"

Session Continuity (CRITICAL)

AI agents lose context between sessions. This section ensures Scrum state survives session boundaries.

Session Start Protocol

On every session start (or /scrum invocation after setup), agents MUST:

1. Read docs/scrum/sprints/current.md -- current sprint state, goal, item statuses
2. Read the Sprint Summary section in current.md for quick context
3. Read docs/scrum/backlog.md -- remaining work and priorities
4. Read docs/scrum/definition-of-done.md -- quality criteria

This gives the agent enough context to continue where the previous session left off.

Sprint Summary Maintenance

The Developer MUST update the Sprint Summary section in current.md whenever:

• An item's status changes
• A significant decision is made
• A blocker is encountered or resolved
• The session is about to end (if known)

Sprint Summary format:

### Sprint Summary (for session continuity)
**What**: {1-2 sentence summary}
**Progress**: {X/Y items done}
**Key decisions**: {recent important decisions, max 3}
**Next action**: {what should happen next if session resumes}

CLAUDE.md Integration

The ## Scrum section in CLAUDE.md (written by /scrum setup or /scrum update) ensures that even without explicit /scrum invocation, new sessions are aware of:

• Artifact locations
• Flow rules
• Anti-patterns

This section is loaded automatically on every session start.

Automatic Ceremony Flow

Ceremonies chain automatically. The user does NOT invoke each one. Agents MUST NOT stop between ceremonies to ask for permission or confirmation.

User expresses desire
  → PO: create backlog item automatically
  → Sprint Planning: auto-start if no active sprint
  → Dev: implement (branch + changes)
  → Increment ready → Sprint Review (present to stakeholder)
  → Stakeholder feedback → Retrospective auto-runs
  → SM improves org → Next sprint auto-starts (if backlog has items)

Flow Rules

1. Never stop between ceremonies. Review → Retro → Next Planning is one continuous flow.
2. Stakeholder feedback is async. Present the increment, ask for feedback, but proceed to Retro immediately. Feedback received later is incorporated in the next Backlog Refinement.
3. Backlog operations need no approval. The Scrum Team autonomously manages the backlog. Update it, reorder it, add items -- no "is this OK?" needed.
4. When backlog has items, start the next sprint. After Retro completes and backlog is not empty, Sprint Planning begins immediately.
5. When backlog is empty, self-generate work. Verify Product Goal achievement, propose technical improvements, or discover new backlog items. Never say "what should I do next?"

The user only needs to:

• Express desires
• Review increments and give feedback (async -- does not block the flow)

Anti-Patterns (NEVER DO)

These patterns caused sessions to stall. They are explicitly forbidden:

Rule: If you catch yourself about to ask permission to continue Scrum flow, STOP and just do it. Rule: If you catch yourself announcing what you'll do next, STOP and do it in this turn instead. Rule: If there is no active Sprint and you are about to write code or make changes, STOP and start Sprint Planning first.

File Structure

Local Scrum records (always created, regardless of tools):

docs/scrum/
  .scrum-version                          # Installed skill version (e.g., "1.0.0")
  definition-of-done.md                   # DoD (evolves through retros)
  sprints/
    current.md                            # Current sprint state
    YYYY-MM-DD_sprint-NNN/                # Sprint archive
      plan.md                             # Sprint Goal + items
      log.md                              # Progress log
      review.md                           # Review record
      retrospective.md                    # Retro record
  logs/
    failures.md                           # Failure log
    decisions.md                          # Design decisions
    adaptations.md                        # Real-time adaptations
    role-interactions.md                  # Cross-role handoff log

If external issue tracker is available, docs/scrum/backlog.md is optional. The external tracker IS the backlog. Local records track the Scrum process itself.

Setup (First-Time /scrum)

When docs/scrum/ does NOT exist:

Step 1: Detect Environment

• Tech stack (pyproject.toml, package.json, Cargo.toml, etc.)
• VCS and remote platform (what tools/skills/MCPs are available?)
• Existing CLAUDE.md and docs/
• Ask stakeholder about preferred tools if unclear

Step 2: Create Local Structure

Read reference files from references/ and adapt them to the project:

Scrum records (docs/scrum/):

• definition-of-done.md ← adapt the {Adapt to project} placeholder in the Testing section based on Step 1 detection results:
  • Python (pyproject.toml + pytest): "pytest passes", and if ruff detected: "ruff check passes"
  • Node.js (package.json + jest/vitest): "npm test passes", and if eslint detected: "eslint passes"
  • Rust (Cargo.toml): "cargo test passes", and if clippy detected: "cargo clippy passes"
  • No test framework detected: "Manual verification documented in review.md"
  • Replace the entire {Adapt to project ...} block with the concrete check items. Do not leave the placeholder.
• sprints/current.md ← empty sprint template
• logs/failures.md ← empty with header
• logs/decisions.md ← empty with header
• logs/adaptations.md ← empty with header
• logs/role-interactions.md ← cross-role handoff log (from references/templates/role-interactions.md)

If no external issue tracker: also create backlog.md from template.

Agents (.claude/agents/) -- adapt, don't just copy:

• scrum-product-owner.md ← adapt artifact locations to the environment
• scrum-master.md
• scrum-developer.md

Agent definitions contain artifact references (backlog location, etc.). When the environment uses external tools, update these references so agents know where to find and manage artifacts.

Rules (.claude/rules/):

• scrum-principles.md
• scrum-values.md
• scrum-role-separation.md

Step 3: Configure Environment Mapping

Record detected tools and how Scrum concepts map to them in docs/scrum/sprints/current.md or CLAUDE.md. Example:

## Scrum Environment
- Backlog: GitHub Issues (via `gh` CLI)
- Sprint tracking: GitHub labels
- Code review: Pull Requests

Or:

## Scrum Environment
- Backlog: docs/scrum/backlog.md
- Sprint tracking: docs/scrum/sprints/current.md
- Code review: Direct stakeholder review

Step 4: Record Skill Version

Write the current skill version to docs/scrum/.scrum-version:

1.0.0

This file is a single line containing only the version number. It is used by the Version Check to detect when the skill has been updated.

Step 5: Update CLAUDE.md

Append ## Scrum (v{version}) section to CLAUDE.md. This section MUST include:

• Artifact locations (backlog, sprint, DoD, logs)
• Ceremony Auto-Flow diagram
• Flow Rules (all 5)
• Anti-Patterns table (all entries)

This is critical. Without this section in CLAUDE.md, new sessions won't load Scrum rules automatically. The /scrum skill is only invoked on demand -- CLAUDE.md is loaded on every session start.

Step 6: Ask for Product Goal

"Scrum を導入しました。このプロジェクトで何を実現したいですか？"

Step 7: Auto-flow

PO agent → create backlog → Sprint Planning → Dev starts.

Status (/scrum status)

1. Read docs/scrum/sprints/current.md for sprint state
2. If external tracker available: query backlog and sprint items
3. Count sprint archives in docs/scrum/sprints/
4. Display in Japanese using this format:

Output format:

## {Project Name} Scrum Status

**Current Sprint**: Sprint {N} -- {Goal}
**Progress**: {completed}/{total} items done

| Item | Status | What It Delivers |
|------|--------|------------------|
| {name} | {status} | {user-facing value} |

**Backlog**: {N} items remaining
**Completed Sprints**: {N} (Sprint 1-{N} archived)

{If no active sprint: "No active sprint. Ready for next Sprint Planning."}

Focus on what the stakeholder cares about: what's being worked on, what they'll get, and what's next.

Update (/scrum update)

Update project Scrum files to match the latest skill version.

When (Auto-detect)

On every /scrum invocation (any argument), the Version Check (above) compares versions:

1. Read skill version from this SKILL.md's metadata.version
2. Read project version from docs/scrum/.scrum-version
3. If versions differ: show message "スキルバージョンが更新されています (v{old} -> v{new})。/scrum update で更新できます。"

The update is NOT automatic -- the user must explicitly run /scrum update.

Process

1. Read versions: Compare metadata.version with docs/scrum/.scrum-version
2. Identify changes: List files that differ between templates and project
3. Update managed files: For each Scrum-managed file:
   • Read the current project file
   • Read the latest template from references/
   • Merge: keep project-specific content, update template-managed sections
   • Files to check:
     • .claude/agents/scrum-*.md <-- references/agents/
     • .claude/rules/scrum-*.md <-- references/rules/
     • docs/scrum/definition-of-done.md <-- references/templates/definition-of-done.md
4. Update CLAUDE.md: Append or update the ## Scrum section in the project's CLAUDE.md. This section MUST include: artifact locations, Flow Rules, and Anti-Patterns. Without this, new sessions won't know about Scrum rules unless /scrum is explicitly invoked.
   • If ## Scrum section exists: replace it with the latest version
   • If ## Scrum section does not exist: append it at the end
   • Never touch non-Scrum content in CLAUDE.md
5. Update version: Write new version to docs/scrum/.scrum-version
6. Report: Show what was updated in Japanese

Customization Preservation

Principle: Never overwrite project-specific adaptations.

Strategy:

• Agent definitions: Update template sections (Role Boundary, Artifacts, Record Format). Preserve project-added sections (custom workflows, project-specific notes). When in doubt, show a diff to the user rather than overwriting.
• Rules: Replace entirely (rules are skill-defined, not project-customized). Important: If a project's SM or Dev adds valuable content to rules files (e.g., Scrum Guide sections, enforcement details), those improvements MUST be upstreamed into the template files in references/rules/ BEFORE the next version bump. Otherwise, /scrum update will delete them. The SM should check for rule-file diffs during Retrospective and upstream any valuable additions.
• DoD: Update Scrum section only. Preserve Quality, Testing, Transparency sections (these are project-adapted by the team).
• CLAUDE.md: Append/update Scrum section only. Never touch non-Scrum content.

Merge Strategy Details

For agent definitions and DoD, use this merge approach:

1. Identify template-managed sections: Sections that originate from the skill templates (e.g., "Role Boundary", "Artifacts" in agent definitions; "Scrum" section in DoD).
2. Identify project-specific sections: Any sections or content added by the project team that are NOT in the template.
3. Merge:
   • Replace template-managed sections with the latest template content
   • Keep project-specific sections intact at their current location
   • If a section exists in both template and project with different content, prefer the template version for template-managed sections
4. Report changes: List each file and what was updated vs. preserved

Version File

docs/scrum/.scrum-version format:

1.0.0

Single line, just the version number. Created during /scrum setup (Step 4), updated by /scrum update.

Sprint Archival

Executor: SM agent -- after Retrospective Step 4 completes.

1. Determine sprint number: Count existing sprint-NNN directories in docs/scrum/sprints/, add 1, zero-pad to 3 digits
2. Create docs/scrum/sprints/YYYY-MM-DD_sprint-NNN/ (date = retrospective execution date)
3. Save: plan.md, log.md, review.md, retrospective.md (extracted from current.md and retrospective output)
4. Reset docs/scrum/sprints/current.md to: No active sprint. Backlog has items -- ready for Sprint Planning.
5. If external tracker: update item statuses (close completed items, etc.)
6. Commit the archive

Logging

All events logged with timestamps in docs/scrum/logs/: