parallel-orchestrator

name	parallel-orchestrator
description	ARCHIVED — Use orchestrator-directives-skill instead. Orchestrate parallel agent workflows using Task tool. Activate when planning multi-agent work or coordinating concurrent feature implementation.

Parallel Orchestrator Skill (ARCHIVED)

This skill is archived. Use /wipnote:orchestrator-directives-skill for current orchestration patterns.

Core Principle: 6-Phase Parallel Workflow

1. ANALYZE   → wipnote analytics summary + wipnote find features --status todo
2. PREPARE   → Cache shared context, isolate tasks
3. DISPATCH  → Spawn agents in ONE message (parallel)
4. MONITOR   → Track health metrics per agent
5. AGGREGATE → Collect results, detect conflicts
6. VALIDATE  → uv run pytest && uv run ruff check

---

Phase 1: Pre-Flight Analysis

# Check what can be parallelized
wipnote analytics summary
wipnote find features --status todo
wipnote analytics summary

Decision Criteria

Condition	Action
2+ independent todo features	Can parallelize
Shared file edits	Partition or sequence
All features blocked	Resolve bottlenecks first

---

Phase 2: Context Preparation

Identify files all agents need and share that context in each agent's prompt. Partition file ownership to avoid conflicts.

---

Phase 3: Dispatch with Task Tool

CRITICAL: Send ALL Task calls in a SINGLE message for true parallelism!

# CORRECT: All in one message (parallel)
Task(subagent_type="wipnote:gemini-operator", description="Research API", prompt="...")
Task(subagent_type="wipnote:sonnet-coder", description="Implement feat-123", prompt="...")
Task(subagent_type="wipnote:sonnet-coder", description="Implement feat-456", prompt="...")

# WRONG: Sequential messages (not parallel)
# result1 = Task(...)  # Wait for completion
# result2 = Task(...)  # Then next one

---

Phase 4: Monitor (During Execution)

Agents track their own health via transcript analytics.

Healthy Patterns

Pattern	Why
Grep → Read	Search before reading
Read → Edit → Bash	Read, modify, test
Glob → Read	Find files first

---

Phase 5: Aggregate Results

After all agents complete, collect their results and check for conflicts via git diff / test suite.

uv run pytest
uv run ruff check --fix
git diff --stat

---

Phase 6: Validate

# Verify all tests pass
uv run pytest

# Commit unified changes
wipnote feature complete <feat-id>

---

When NOT to Parallelize

Situation	Reason	Alternative
Shared dependencies	Conflicts	Sequential + handoff
Tasks < 1 minute	Overhead not worth it	Sequential
Overlapping files	Merge conflicts	Partition files

---

Troubleshooting

"Not enough independent tasks"

# Check what's available
wipnote analytics summary
wipnote find features --status todo

File conflicts detected

• Improve task isolation in Phase 2
• Consider sequential execution for overlapping work