Get the kit — $39

Your Claude Code agents should be getting better while you sleep on it.

AgentCycle is a working multi-agent system for Claude Code — Planner, Coder, Reviewer, Designer — plus an Improver agent that reads real metrics from your sessions and tunes the other agents every 3 cycles. Extracted from daily production use. Not a prompt pack.

Get the kit — $39 Launch price · $59 after the first 50 copies
// the loop Planner writes the PRD + failing tests tests ARE the spec Coder makes tests green can't touch the spec Reviewer ships or rejects + logs metrics every verdict counted × 3 cycles, then a hook fires… Improver reads the metrics, finds repeat failures, proposes evidence-based diffs to the agents themselves → approved by you, committed to git, reversible

What's in the box

5× agent definitions

Planner, Coder, Reviewer, Designer, Improver — with strict role boundaries so handoffs stay clean. The Planner can't code. The Coder can't change specs.

1× cycle engine

agent-cycle.sh + a PostToolUse hook: counts Reviewer finishes, logs a metrics row per verdict, and nudges Claude when the Improver is due. Every invocation logs — never silent.

1× CLAUDE.md template

Drop-in project conventions that wire the flow into any repo: docs layout, handoff rules, when to spawn which agent.

5-min setup guide

Copy two folders, merge one hook, verify with three commands. Works on macOS and Linux with stock bash + python3.

Why this beats writing your own agents

Anyone can write a "you are a senior engineer" prompt. The hard part is the part nobody ships: the feedback loop. Most agent setups degrade — the same import-path bug, the same spec ambiguity, the same review rejection, forever — because nothing closes the loop.

AgentCycle closes it. Every Reviewer verdict logs rejections, iterations, and issues found. The Improver reads that table, finds patterns with at least two data points, and proposes minimal diffs to the agent files — with evidence, behind your approval, committed to git.

The version you install today is the worst it will ever be.

Get AgentCycle

The complete kit

Instant download · all 5 agents + cycle engine + setup guide
Use on unlimited personal projects · free updates to v1.x

$39 $59
Buy now

Questions

Do I need anything besides Claude Code?

No. macOS or Linux with bash and python3 (stock on both). Git recommended so the Improver's changes are reversible.

Will this work with my existing CLAUDE.md and agents?

Yes — the agents install as separate files in ~/.claude/agents/ and the project template appends to your CLAUDE.md. Nothing is overwritten without you choosing to.

How is this different from a prompt pack?

Prompt packs are static text. AgentCycle ships an executable feedback loop: a hook-driven counter, a metrics log, and a meta-agent whose whole job is editing the other agents based on your real failure data.

What do I get exactly, and how?

A ZIP delivered immediately after checkout: 5 agent .md files, agent-cycle.sh, hook config, CLAUDE.md template, README and SETUP guide.

Refunds?

If it doesn't work for you within 14 days, email and you get your money back. No forms.