Plan Backlog - AI Scrum Master

Transform rough notes into a well-organized, parallelizable backlog with worker-ready prompts.

Steps

Add these to your to-dos:

1. Analyze current state - Get backlog stats and ready/blocked issues
2. Break down epics - Use /planner:breakdown skill (Sonnet) for decomposition
3. Prioritize and wire dependencies - Set priorities, add blockers
4. Assign quality gates - Create gate issues based on issue characteristics
5. Prepare issue prompts - Use /prompt-writer:write for each backlog issue (Haiku)
6. Output sprint plan - Present waves of parallelizable work

Step 1: Analyze Current State

Using MCP (preferred):

mcp__beads__stats()                    # Overview
mcp__beads__ready()                    # What's unblocked
mcp__beads__blocked()                  # What's stuck
mcp__beads__list(status="open")        # All open work

CLI fallback:

bd stats
bd ready --json
bd blocked --json
bd list --status open --json

Step 2: Break Down Large Work (Epics)

Use the /planner:breakdown skill for strategic decomposition of epics.

The skill runs with Sonnet and will:

• Apply decomposition patterns (feature, refactor, or bug fix)
• Identify task sizing (target S-M tasks)
• Detect file overlaps that need sequencing
• Suggest wave organization

Using MCP:

# Create epic
mcp__beads__create(
  title="Auth System",
  issue_type="epic",
  priority=1
)

# Add subtasks
mcp__beads__create(title="Design auth flow", issue_type="task")
mcp__beads__create(title="Implement login", issue_type="task")
mcp__beads__create(title="Add tests", issue_type="task")

# Wire dependencies
mcp__beads__dep(issue_id="IMPL-ID", depends_on_id="DESIGN-ID")
mcp__beads__dep(issue_id="TESTS-ID", depends_on_id="IMPL-ID")

Step 3: Prioritize and Wire Dependencies

For each issue, consider:

• Is it blocking other work? -> Raise priority
• Is it a quick win? -> Raise priority
• Does it have dependencies? -> Add them
• What labels apply? -> Add them

Using MCP:

# Set priority (0=critical, 1=high, 2=medium, 3=low, 4=backlog)
mcp__beads__update(issue_id="ID", priority=1)

# Add dependencies (blocker blocks blocked)
mcp__beads__dep(issue_id="BLOCKED-ID", depends_on_id="BLOCKER-ID")

CLI fallback:

bd update ID --priority 1 --json
bd dep add BLOCKED-ID BLOCKER-ID --json
bd label add ID frontend,auth --json

Step 4: Assign Quality Gates

Based on issue characteristics, assign appropriate quality checkpoints. The /conductor:gate-runner (or the one-shot finalize-issue.sh) runs these checkpoints before merging.

Gate Types

Gate	Purpose	Checkpoint Skill
codex-review	Code review via Codex	/conductor:reviewing-code
test-runner	Run project tests	/conductor:running-tests
visual-qa	Visual/UI verification	/conductor:visual-qa
docs-check	Changelog/docs hygiene	/conductor:docs-check

Assignment Heuristics

Analyze each issue and suggest gates based on:

Indicator	Suggested Gates
Issue Type
Bug fix	codex-review
New feature	codex-review, test-runner
Refactor	codex-review, test-runner
Chore/config	(none or codex-review)
Docs only	(none)
Epic close	codex-review
Files Touched
*.test.ts, *.spec.ts	test-runner
README.md, docs/, *.md	docs-check
Labels
needs-tests, testing	test-runner
needs-review, security	codex-review
needs-docs	docs-check

Visual QA - Conductor Only (Not for Workers)

IMPORTANT: visual-qa should NOT be assigned as a gate label on work issues.

Visual QA requires:

• Changes merged to main (extension rebuilt)
• Browser access (workers on worktrees don't have isolated browsers)
• Tab group isolation (which may be disabled)

Instead, the conductor runs visual QA after merging:

# After merge, conductor can run visual-qa directly
/conductor:visual-qa

If an issue truly needs visual verification, add a note in the issue description for the conductor to check post-merge, but don't add gate:visual-qa as a label.

Assigning Gates (Recommended: labels on the work issue)

After determining which gates apply, add labels to the work issue:

Using CLI:

# Add one gate
bd label add ISSUE-ID gate:codex-review

# Add multiple gates
bd label add ISSUE-ID gate:codex-review,gate:test-runner,gate:visual-qa,gate:docs-check

Using MCP:

mcp__beads__update(issue_id="ISSUE-ID", labels=["gate:codex-review", "gate:test-runner"])

Docs-check bypass label (explicit, searchable):

bd label add ISSUE-ID no-changelog

Presenting Gate Suggestions

When outputting the sprint plan, show suggested gates:

## Issue Analysis

| Issue | Type | Files | Suggested Gates |
|-------|------|-------|-----------------|
| bd-xxx | bug | Terminal.tsx | codex-review, visual-qa |
| bd-yyy | feature | api.js, api.test.js | codex-review, test-runner |
| bd-zzz | docs | README.md | (none) |

**Create these gates?** [y/N]

User Override

Always allow the user to:

• Add additional gates
• Remove suggested gates
• Skip gates entirely for low-risk changes

Would you like to:
1. Create all suggested gates
2. Select gates per issue
3. Skip gate assignment

Step 5: Prepare Issue Prompts

Use /prompt-writer:write for each backlog issue to craft worker-ready prompts.

The prompt-writer skill runs with Haiku and will:

• Read issue details
• Discover relevant skills
• Explore key files (3-5 only)
• Craft a focused prompt following Claude 4 best practices
• Store prompt in issue notes
• Mark issue as ready

For batch processing, use /prompt-writer:write-all to process all backlog issues in parallel.

Step 6: Output Sprint Plan

Before presenting the plan, verify prompts are written:

ready = mcp__beads__ready()
missing_prompts = []
for issue in ready:
    full = mcp__beads__show(issue_id=issue['id'])
    notes = full[0].get('notes', '') if full else ''
    if '## prepared.prompt' not in notes and 'prepared.prompt:' not in notes:
        missing_prompts.append(issue['id'])

if missing_prompts:
    print(f"⚠ {len(missing_prompts)} issues need prompts: {missing_prompts}")
    print("Run /prompt-writer:write-all first")
    # DO NOT proceed to spawn offer

Only after all prompts are ready, present the organized backlog:

## Wave 1 (Ready Now - All Prompts Written ✓)
| Issue | Priority | Type | Description |
|-------|----------|------|-------------|
| bd-xxx | P1 | bug | Fix login redirect |
| bd-yyy | P2 | feature | Add dark mode toggle |

## Wave 2 (After Wave 1)
| Issue | Blocked By | Description |
|-------|------------|-------------|
| bd-zzz | bd-xxx | Refactor auth flow |

Ready to spawn workers on Wave 1?

Decision Guidance

Situation	Action
Blocks 3+ issues	Priority 0-1
Quick win (<1hr)	Priority 1-2
User-facing bug	Priority 0-1
Nice-to-have	Priority 3-4
Large feature	Break into epic + subtasks

Sparse Backlog?

If there are few/no ready issues, offer to brainstorm:

"Your backlog is light. Want to brainstorm what needs to be done?"

Then help the user think through:

• What work needs to be done (rough ideas -> concrete tasks)
• How to structure it (epics, dependencies, waves)
• What would "done" look like

See the brainstorm skill references for dependency patterns and epic structures.

Start by running mcp__beads__stats() and mcp__beads__ready() to understand the current state.

Checklist Before Offering to Spawn

Do NOT ask "Ready to spawn?" until ALL of these are true:

• All issues have priorities set
• Dependencies are wired
• Gate labels assigned (or explicitly skipped)
• All ready issues have prepared.prompt in notes ← Most commonly missed!

If prompts are missing, run /prompt-writer:write-all first.