wbs-orchestration

name: wbs-orchestration description: | WBS-driven work sessions and parallel agent orchestration. Modes: STATUS (show progress), NEXT (dispatch agents), FINISH (close epic), BLAST (max parallel). triggers: - "wbs" - "epic" - "next.*task" - "parallel.*agent" - "blast" - "finish.*epic" - "wave.*dispatch" - "work session" - "v3 architecture"

WBS Orchestration Skill

Purpose

Unified WBS (Work Breakdown Structure) management:

1. Session tracking — Progress display, task claiming, session logging
2. Agent orchestration — Parallel agent dispatch in coordinated waves

Mode Detection

Session Start Protocol (MODE: STATUS)

Step 1: Read Progress Tracker

cat .claude/plans/enumerated-crafting-otter.md

Step 2: Parse Current State

• Find epic with 🎯 marker (current focus)
• Count ✅ vs ⏳ tasks
• Calculate progress percentage

Step 3: Display Status

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📋 WBS STATUS — V3 Service Architecture
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Progress: XX% (XX/267 SP)
Current Epic: N — [Epic Name]

🎯 NEXT TASKS:
  • [Task ID] — [Description]
  • [Task ID] — [Description]

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Step 4: Create Session Tasks

Use TaskCreate for each 🎯/⏳ task in current epic.

Agent Dispatch Protocol (MODE: NEXT/BLAST)

Step 1: Context Load

1. TaskList → Get current task status
2. Read WBS tracker → epic requirements
3. Identify: completed, in_progress, pending, blocked

Step 2: Wave Dispatch

WAVE 1 (Foundation) — Parallel
├─ kraken-<feature-a>  → Task #X
├─ kraken-<feature-b>  → Task #Y
└─ kraken-<feature-c>  → Task #Z

↓ Wait for Wave 1 completion

WAVE 2 (Integration) — Parallel
├─ kraken   → Service refactoring
├─ arbiter  → Integration tests
└─ spark    → TypeScript error fixes

Key: Dispatch ALL agents in a SINGLE message for parallel execution.

Agent Selection Matrix

Finish Protocol (MODE: FINISH)

1. Run `bunx tsc --noEmit` — fix any errors
2. Run tests for current epic
3. Update all in_progress → completed
4. Update tracker: 🔄 → ✅
5. Generate deliverables table
6. Append to session log

Session End Protocol

Step 1: Update Tracker

Edit .claude/plans/enumerated-crafting-otter.md:

• Change ⏳ → ✅ for completed tasks
• Change 🔄 → ✅ for finished in-progress

Step 2: Log Session

Append to .claude/plans/wbs-session-log.md:

## YYYY-MM-DD — Session [short-id]
**Agent:** [name]
**Epic:** [N] ([Name])
**Completed:** [task IDs]
**Next:** [next task ID + description]
**Blockers:** [any blockers or "None"]

Step 3: Identify Next

Find next 🎯 task for future sessions.

Files

Progress Symbols

Task Status Protocol

// On agent dispatch
TaskUpdate({ taskId, status: 'in_progress' })

// On agent completion
TaskUpdate({ taskId, status: 'completed' })

// On blocker
TaskUpdate({ taskId, status: 'pending', description: 'BLOCKED: <reason>' })

Quality Gates

Before reporting epic complete:

• bunx tsc --noEmit passes (0 errors)
• All tests pass for affected files
• All task statuses updated to completed
• Tracker markdown updated (🔄 → ✅)
• Session log appended
• Deliverables table generated

Agent Dispatch Template

Each agent MUST receive:

## Task #N: <Task Title>

You are <agent-type>, a <role-description>.

### Research First
- Use **deepwiki** MCP for @Effect-TS/effect patterns
- Use **effect-docs** MCP for API reference
- Invoke **/<domain-skill>** for context

### Requirements
[Specific task requirements from WBS]

### TDD Workflow
1. Write failing test
2. Implement minimal code
3. Verify test passes
4. Refactor if needed

### Verification
- Run `bunx tsc --noEmit` before completion
- All tests must pass

### Report
Write to `.claude/cache/agents/<type>/latest-output.md`

Decision Tree

User invokes /wbs or /wbs-orchestration
│
├─ Parse mode from user message
│   ├─ "status/progress"     → MODE: STATUS
│   ├─ "finish/complete"     → MODE: FINISH
│   ├─ "blast/parallel"      → MODE: BLAST
│   ├─ "claim <task>"        → MODE: CLAIM
│   ├─ "complete <task>"     → MODE: COMPLETE
│   ├─ "log"                 → MODE: LOG
│   └─ "next/go/continue"    → MODE: NEXT
│
├─ Load context
│   ├─ Read tracker → enumerated-crafting-otter.md
│   └─ TaskList → current task status
│
├─ Execute mode
│   ├─ STATUS   → Display progress, suggest tasks
│   ├─ NEXT     → Wave 1 → Wave 2 agent dispatch
│   ├─ FINISH   → Fix errors, tests, close tasks
│   ├─ BLAST    → Max parallel agents
│   ├─ CLAIM    → Mark task 🔄 in tracker
│   ├─ COMPLETE → Mark task ✅ in tracker
│   └─ LOG      → Show session log
│
└─ Report
    ├─ Progress percentage
    ├─ Test counts
    ├─ Deliverables table
    └─ Next epic preview

Multi-Agent Coordination Rules

Recovery Protocol

If context compacts:

1. TaskList → identify last known state
2. Read `.claude/cache/agents/*/latest-output.md`
3. Read tracker → find 🔄 tasks
4. Resume from next pending task

Sub-Skills to Invoke

Example Invocations

Show Status

User: "/wbs"
Claude: [Reads tracker, displays progress, suggests next tasks]

Continue WBS

User: "next"
Claude: [Loads TaskList, identifies pending, dispatches Wave 1]

Finish Epic

User: "/wbs finish"
Claude: [Runs tsc, fixes errors, runs tests, closes tasks, logs session]

Maximum Parallel

User: "blast all pending tasks"
Claude: [Dispatches 5 parallel agents across epics]

Claim Task

User: "/wbs claim EL-2.3"
Claude: [Marks EL-2.3 as 🔄 in tracker, creates TaskCreate entry]

Related Skills