Agentic AI

Agents

An agent is a specialist Claude Code can hand work to — a reviewer, an analyst, a tester. Each has one job and clear limits.

  • Agent

    Delivery Advisor

    Tells you the single most valuable next step on a goal or project, in best-practice order, and flags the decisions only you can make. It recommends; it never builds.

    Best for: Starting a new goal and not knowing the order of steps

  • Agent

    Requirements Analyst

    Turns a vague request into clear, numbered requirements with acceptance criteria — and lists every question you still need to answer. It writes requirements, never code.

    Best for: Founders and product owners with ideas but no spec

  • Agent

    Test Runner

    Runs your project's checks after a change and tells you honestly what passed, what failed and why — and what it did not run. It never edits your files.

    Best for: Validating a change before you call it done

  • Agent

    Change Reviewer

    Reviews a change against your own project rules before you commit it: data-model changes, new dependencies, persistence, scope, security and traceability. Read-only.

    Best for: A second pair of eyes on every change before it is committed

  • Agent

    Prompt Selector

    Reads a request, picks the prompt-engineering technique that gives the best accuracy for the lowest cost, and rewrites the prompt for you — without adding facts you did not give.

    Best for: Getting more reliable answers from the same model

Also see the Skills.