🧵 Agents Threads

A Claude Code plugin providing 7 specialized execution patterns (threads) for different types of software engineering tasks. Each thread implements a distinct workflow optimized for specific scenarios.

Why Threads?

When working with AI coding assistants, different tasks require different approaches:

• Simple bug fixes need quick iteration loops
• Complex migrations need phased approval checkpoints
• Design decisions benefit from parallel exploration
• Large features need task breakdown and tracking

Agents Threads gives you precise control over how Claude approaches your task, not just what it does.

Installation

Option 1: Clone and Link

git clone https://github.com/YOUR_USERNAME/agents-threads.git
# Add the path to your Claude Code plugin settings

Option 2: Direct Plugin Reference

Add to your Claude Code settings:

{
  "plugins": ["path/to/agents-threads"]
}

Quick Reference

Thread	Trigger Phrase	Best For
Base	Run a Base Thread on...	Standard tasks
P-Thread	Start a P-Thread [N] for...	Exploring options
C-Thread	Start a C-Thread for...	High-stakes changes
F-Thread	Run an F-Thread on...	Optimal solutions
B-Thread	Start a B-Thread for...	Large projects
L-Thread	Run an L-Thread to fix...	Debugging loops
Z-Thread	Execute Z-Thread: [task]	Full autonomy

The 7 Threads

🔵 Base Thread — Standard Execution

Trigger: "Run a Base Thread on [task]"

The foundation pattern. Simple three-phase execution:

1. Plan — State the approach
2. Execute — Do the work
3. Verify — Confirm it works

Use when: You have a straightforward task that doesn't need special handling.

---

🟣 P-Thread — Parallel Exploration

Trigger: "Start a P-Thread [count] for [task]"

Explores multiple solutions simultaneously in isolated folders.

Use when:

• You're unsure which approach is best
• You want to compare different architectures
• Design decisions have trade-offs you want to evaluate

Example:

Start a P-Thread 3 for implementing the caching layer

Creates exploration_1/, exploration_2/, exploration_3/ with different strategies.

---

🟡 C-Thread — Chained Phases

Trigger: "Start a C-Thread for [task]"

Sequential phases with mandatory user approval between each:

1. Plan → approval required
2. Scaffold → approval required
3. Implement → approval required
4. Cleanup

Use when:

• Database migrations
• Breaking API changes
• Security-sensitive modifications
• Any change where mistakes are costly

---

🟠 F-Thread — Fusion

Trigger: "Run an F-Thread on [task]"

Generates three variants optimized for different qualities, then combines the best of each:

Variant	Focus
A	Performance & Speed
B	Readability & Clean Code
C	Robustness & Error Handling

Use when:

• Writing critical functions
• Code that needs to be both fast AND maintainable
• You want the "best of all worlds"

---

🔴 B-Thread — Big/Manager

Trigger: "Start a B-Thread for [objective]"

Breaks large objectives into tracked tasks via TASKS.todo file:

# TASKS.todo - Implement Auth System

[x] 1. Create user model
[x] 2. Add password hashing
[ ] 3. Implement JWT tokens  <-- in progress
[ ] 4. Build login endpoint
[ ] 5. Add session management
[ ] 6. Write tests

Use when:

• Multi-day features
• Complex refactoring
• Any work spanning multiple files/systems

---

🟢 L-Thread — Ralph Loop

Trigger: "Run an L-Thread to fix [target]. Use [command] as verification."

A self-correcting loop that doesn't stop until tests pass:

┌──────────────────────────┐
│  1. Attempt fix          │
│         ↓                │
│  2. Run verification     │
│         ↓                │
│  3. Pass? → Done         │
│     Fail? → Back to 1    │
└──────────────────────────┘

Safety: Stops after 5 consecutive failures to ask for help.

Use when:

• Stubborn bugs
• Test failures you want fixed automatically
• "Just make it work" scenarios

Example:

Run an L-Thread to fix auth.ts. Use `npm test auth` as verification.

---

⚫ Z-Thread — Zero-Touch Autonomy

Trigger: "Execute Z-Thread: [task]"

Full autonomous mode with git safety:

1. Creates isolated branch: auto/z-thread-[timestamp]
2. Implements feature independently
3. Runs all tests
4. Only commits if tests pass (reverts otherwise)

Use when:

• You trust the AI's judgment
• Task is well-defined
• You want hands-off execution
• Git safety net is acceptable

Requires: Initialized git repository

---

Choosing the Right Thread

Is this a simple, well-defined task?
├─ Yes → Base Thread
└─ No ↓

Do you need to explore multiple approaches?
├─ Yes → P-Thread
└─ No ↓

Is this high-stakes / needs approval gates?
├─ Yes → C-Thread
└─ No ↓

Do you need optimal code quality?
├─ Yes → F-Thread
└─ No ↓

Is this a large, multi-part objective?
├─ Yes → B-Thread
└─ No ↓

Are you fixing a bug with clear test criteria?
├─ Yes → L-Thread
└─ No ↓

Do you want full autonomy with git safety?
├─ Yes → Z-Thread
└─ No → Base Thread

File Structure

agents-threads/
├── plugin.json              # Plugin manifest
├── README.md                # This file
├── MANUAL.md                # Detailed user manual
├── Threads.md               # Original specifications
├── CLAUDE.md                # Claude Code guidance
└── .claude/
    └── skills/
        ├── base-thread.md
        ├── p-thread.md
        ├── c-thread.md
        ├── f-thread.md
        ├── b-thread.md
        ├── l-thread.md
        └── z-thread.md

Contributing

Contributions welcome! Ideas for new thread types or improvements to existing ones can be submitted as issues or PRs.

License

MIT License - Use freely in your projects.