Anabranch

🌿 Anabranch

**Anabranch is an experimental orchestration layer for autonomous ticket execution.**

Instead of manually prompting an IDE agent for every task, Anabranch monitors your ticketing system, evaluates ticket complexity, and attempts to automatically open pull requests for low-complexity work — asynchronously.

The goal is not to replace coding agents or MCP integrations. The goal is to explore whether the “boring majority” of tickets can be automated reliably without requiring a developer to initiate each interaction.

Anabranch sits above existing AI coding agents and focuses on orchestration, gating, and workflow automation.

---

The Problem

Modern AI coding agents are powerful, but they are still largely **manual**:

1. A ticket appears.
2. A developer opens their IDE.
3. They prompt an agent.
4. They wait.
5. They iterate.
6. They open a PR.
7. A second developer reviews the PR.

This works — but it does not change the interaction model.

For repetitive, well-scoped tickets, this loop may be unnecessary.

Anabranch explores a different model:

What if low-complexity tickets could be processed automatically, and developers only stepped in for review and guidance?

---

What Anabranch Does

When a ticket is created or updated:

1. Ticket Monitoring

• Listens to Jira via webhooks.

2. AI-Based Assessment

• Evaluates:
  • Scope (trivial → huge)
  • Confidence level
  • Affected repositories

3. Smart Gating

• Proceeds only when confidence and scope thresholds are met.
• Skips ambiguous or high-risk tasks.

4. Autonomous Execution

• Creates isolated git worktrees.
• Delegates implementation to a supported AI coding agent.
• Runs tests.
• Opens a pull request.

5. Human Review

• Developers review, comment, and guide.
• Iteration happens via standard PR workflows.

The interaction becomes asynchronous by default: Developers review outcomes rather than initiate execution.

---

Getting Started

1. Setup GitHub App

Anab