Thread-Based Engineering Patterns

This skill provides guidance on choosing and implementing the right thread pattern for your task.

Quick Reference

Thread	Best For	Complexity	Human Time
Base	Simple tasks	Low	High
Parallel	Independent work	Medium	Medium
Chained	Critical work	Medium	High
Fusion	High confidence	Medium	Low
Big	Complex orchestration	High	Low
Long	Extended autonomy	High	Very Low
Zero-Touch	Full automation	Very High	None

Decision Tree

Is it a simple, single task?
├─ YES → Base Thread
└─ NO → Can tasks run independently?
         ├─ YES → Need multiple perspectives?
         │        ├─ YES → Fusion Thread
         │        └─ NO → Parallel Thread
         └─ NO → Is it high-risk/production?
                  ├─ YES → Chained Thread
                  └─ NO → Need specialized agents?
                           ├─ YES → Big Thread
                           └─ NO → Want extended autonomy?
                                    ├─ YES → Have validation suite?
                                    │        ├─ YES → Zero-Touch or Long
                                    │        └─ NO → Long Thread
                                    └─ NO → Base Thread

Pattern Details

Base Thread

When: Simple tasks, learning, experimentation Pattern: Prompt → Tools → Review Example: "Explain this codebase"

Parallel Thread (P)

When: Multiple independent tasks, scaling compute Pattern: Spawn N agents, each on different work Example: Review auth/, api/, and db/ simultaneously

Chained Thread (C)

When: Production work, large migrations, high-risk changes Pattern: Phase 1 → Review → Phase 2 → Review → ... Example: Plan migration → Review → Execute → Review → Validate

Fusion Thread (F)

When: Need high confidence, comparing approaches Pattern: N agents same prompt → Aggregate best Example: 3 reviewers → Synthesize findings

Big Thread (B)

When: Complex workflows needing specialized agents Pattern: Primary agent orchestrates sub-agents Example: Researcher → Reviewer → Builder → Validator

Long Thread (L)

When: Extended autonomous work, self-correcting Pattern: Work → Stop Hook validates → Continue if needed Example: "Fix all bugs until tests pass" (Ralph Wiggum)

Zero-Touch Thread (Z)

When: Mature codebase, comprehensive validation Pattern: Prompt → Work → Auto-validate → Done Example: Automated dependency updates with full CI

Improvement Metrics

You're getting better when you:

1. Run more threads (P-Threads)
2. Run longer threads (L-Threads)
3. Run thicker threads (B-Threads)
4. Run fewer checkpoints (toward Z-Threads)

See Also

• examples.md for detailed code examples
• Plugin commands: /agent-threads:<type>
• Scripts in scripts/ directory