tapagents

---

TapAgents is a small, opinionated set of agents that turn Claude Code into a working product team. Instead of one assistant trying to do everything, you get 14 curated agents with specific jobs — a VP of Product, a VP of Engineering, a Designer, an adversarial Critic, a Quality Engineer, plus operational roles. You stay in the CEO seat. They handle the work.

It is deliberately narrow. Where other agent frameworks ship 100+ agents and 300+ tools, TapAgents ships 14 agents and 15 slash commands. Less surface area, more discipline, no swarm topology to understand.

Quick Start

Two install paths, depending on how you work:

Path A — Claude Code marketplace (recommended)

/plugin marketplace add tapintomymind/tap-agents
/plugin install tapagents

That's it. Slash commands appear: /team, /intake, /architect, /designer, /critic, /status, /briefing. Run /team and describe what you want to build.

Path B — npm dependency (programmatic consumers)

npm install @tapintomymind/tap-agents

Use this when you're building tooling on top of TapAgents — e.g. an internal dashboard that scaffolds new projects with the framework baked in. Exposes the agent + command + protocol definitions as a typed JavaScript module plus the raw .md files for filesystem-based access.

The Founding Team

Seven core agents do the load-bearing work. The other seven (db-admin, backlog-curator, ui-ux-reviewer, ops-security, plus stubs) activate when their triggers fire.

Agent	Role	Talks to
Intake	Director of Product Discovery — interviews you, briefs the team	You ⇄ Team
Executive Assistant	Chief of Staff — surfaces decisions, delivers briefings	Team → You
Conductor	CTO/CPO — routes work, enforces state machine, runs consistency checks	Team-internal
Strategist	VP Product — turns briefs into PRDs	Conductor-driven
Architect	VP Engineering — turns PRDs into scope + tech strategy + Tier 2 scaffold	Conductor-driven
Critic	Independent advisor — adversarial review of every artifact	Reads everything
Org Designer	Head of People — watches team performance, proposes splits & new roles	Slow cadence

The team grows over time. agents/_planned/ holds activation-trigger stubs for GTM Strategist, Growth Analyst, Customer Researcher, Test Engineer, and others. They live as stubs until real friction triggers their activation — no agent shows up until the work demands it.

Claude Code with vs. without TapAgents

Capability	Claude Code alone	With TapAgents
Routing requests	You decide which prompts go where	Conductor routes between Strategist, Architect, Designer, Critic based on project phase
Reviewing artifacts	You review your own work	Critic runs adversarial review on every artifact, automatically
Project state	Whatever's in your scrollback	12-phase state machine with 5 hard checkpoints for your approval
Cross-project memory	Per-conversation only	memory/ directory captures patterns, product principles, lessons learned across projects
Per-project teams	One Claude Code instance per repo	Tier 1 HQ generates Tier 2 stack-specific teams (Next.js architect, Python architect, etc.) per project
Decision surfacing	Mixed into long responses	EA dedicated to summarizing what needs your input, sorted by priority

How It Works

You (CEO)  ⇄  Intake  →  Strategist  →  Architect  →  Tier 2 (per-project team)
              + Critic     + Critic       + Critic        ↓
              + EA         + EA           + EA          MVP shipped
                                           ↓               ↓
                                      Conductor        Reportback to HQ
                                           ↓
                                     State Machine
                                           ↓
                                     Org Designer (watches, proposes splits)