# Startup Review Agents

Eight Claude Code subagents that act as a virtual review board for an early-stage startup: they pressure-test specs, pitches, content, pricing, and brand assets before the real world does.

| Agent | Reviews | Persona |
|---|---|---|
| `engineer` | Specs, PRDs, technical decisions | Pragmatic engineer sized to a tiny team |
| `executive` | Updates, business cases, strategy docs | Skeptical exec/board lens |
| `investor` | Decks, metrics, fundraising narrative | Seed partner who sees 500 decks a year |
| `skeptical-customer` | Pitches, onboarding, pricing | Your ICP — the one who's heard it all before |
| `user-researcher` | Interview notes, field research | UX researcher, says-vs-does discipline |
| `social-media-expert` | Posts, captions, content calendars | Strategist who kills stage-inappropriate content |
| `brand-expert` | Anything carrying the brand | Consistency and distinctiveness enforcer |
| `financial-advisor` | Pricing, models, spend decisions | Fractional CFO, diligence-ready numbers |

## Install

Copy the `.md` files into your project's `.claude/agents/` directory (or `~/.claude/agents/` for all projects). Claude Code picks them up automatically; invoke them by name or let Claude route to them based on their descriptions.

## Customize

Each file has a `> **Customize:**` note marking where to insert your company's real context — product, stage, audiences, brand rules, cost structure. The agents work out of the box, but they get dramatically sharper once those sections reference your actual constraints and a real knowledge base (brand guidelines, finance docs, research notes) they can read.

Two principles baked into all eight, worth keeping when you adapt them:

1. **Structured verdicts.** Every agent ends with an explicit verdict (ship/fix/rework, pass/proceed/hold), not open-ended commentary.
2. **Stage discipline.** They challenge vanity metrics, premature scaling, and assumptions dressed as facts — the failure modes that quietly kill early-stage companies.