How it works¶

N45 orchestrates specialized subagents inside your IDE. The flow depends on the size of the request.

The flow¶

flowchart TD
    User([User]) -->|/n45-feat or /n45-fix| Triage[Triage]
    Triage -->|small, clear scope| Quick[Quick path]
    Triage -->|architectural change| Structured[Structured path]
    Quick --> Executor1[Executor agent]
    Structured --> Discovery[Discovery]
    Discovery --> Spec[Spec]
    Spec --> Roadmap[Roadmap]
    Roadmap --> Executor2[Executor + Reviewer agents]
    Executor1 --> Validation[Validation]
    Executor2 --> Validation
    Validation -->|approved| Merge[Merge / PR]

Two paths, one philosophy¶

Quick path — for small, clear changes (a button color, a typo fix, a minor refactor). The triage agent confirms scope, then spawns an executor on a dedicated branch.

Structured path — for new domains, schema changes, refactors, or anything ambiguous. N45 runs full discovery, builds a spec, generates a roadmap of tasks, and executes them in phases — each one reviewed by a dedicated agent before moving on.

The triage decides which path to take based on four conditions. If any one of them fails (architectural impact, schema change, new public contract, or unclear intent), N45 escalates to the structured path automatically.

The non-negotiable rules¶

• N45 never writes code directly. Every change goes through a specialized agent.
• You stay in the loop. N45 stops and asks whenever a real decision is needed.
• Your patterns are respected. On the first run, N45 reads your codebase and records architectural patterns it will follow.

See Building a feature to see the flow in practice.