ScreenSheriff 3000 — bootstrap

Paste-into-Claude-Code starter. The CLAUDE.md below contains the idea spec, agent-readiness sub-scores, suggested tools, and smoke evals — deterministic, no AI hallucination.

# ScreenSheriff 3000

> Generated by [whycantwehaveanagentforthis.com](https://whycantwehaveanagentforthis.com/result/screensheriff-3000-agent-automatically-reads). Roasted, scored, ready to scaffold.

## What you are building

**Problem:** an agent that automatically reads your screenshots and files them into the right project folder

**Verdict:** ALREADY EXISTS — _"You just described Apple's Intelligent Albums, Windows Recall, and Google Photos — but for your chaotic Downloads folder."_

**Summary:** An agent that watches for new screenshots, uses vision AI to understand the content, and automatically moves them into the correct project folder based on what's on screen.

## Agent-readiness score

Overall: **69/100** (band C)

| Dimension | Score | Why |
|---|---|---|
| Memory required | 21/25 | Stateless or single-session — minimal memory layer. |
| Tool count | 11/25 | Crowded market: at least 8 integrations to compete. |
| Policy surface | 15/25 | Mid-size policy surface — define refusal categories before launch. |
| Eval coverage | 22/25 | Established eval pattern — golden datasets and public benchmarks already exist. |

> Worth building, but plan for the long-tail. ScreenSheriff 3000 needs runway, not just speed.

## Suggested tools

- fetch (HTTP GET on a URL allow-list)
- search (Brave / Tavily / Exa for competitor research)
- database (Postgres / Supabase for user state)

## Smoke evals

- The agent introduces itself as "ScreenSheriff 3000" and refuses tasks outside the stated scope.
- Given the canonical problem ("an agent that automatically reads your screenshots and files them into the right"), the agent produces a plan in ≤ 200 tokens.
- When asked "what's different from Windows Recall?", the agent gives a concrete differentiator, not a marketing line.
- When asked about Apple's threat, the agent acknowledges the risk honestly.
- No private personal data appears in any output (PII redaction smoke test).

## Stack

- Model: `claude-sonnet-4-6` (Anthropic). Override via `ANTHROPIC_MODEL` env.
- Suggested stack: `Electron or Tauri`, `GPT-4o Vision API`, `watchdog (Python) or chokidar (Node)`, `SQLite for folder-rule memory`
- Solo build estimate: 3-5 days for a working prototype, 3 months to make it not embarrassing

## Kill prediction

Apple could obsolete this in 12-18 months. macOS 16 ships a native 'Smart Screenshots' feature in Finder that does exactly this, opt-in, on-device, with zero privacy drama. It'll be a WWDC slide. You'll watch it live.

**Survival strategy:** Go vertical — build exclusively for one profession (e.g., lawyers filing court screenshots, QA engineers filing bug screenshots) with domain-specific folder logic they'd never get from Apple.

## Hand-off

- Read the full analysis: https://whycantwehaveanagentforthis.com/result/screensheriff-3000-agent-automatically-reads
- Open in Anthropic Managed Agents: see the deeplink on the result page
- Claim this idea: https://whycantwehaveanagentforthis.com/result/screensheriff-3000-agent-automatically-reads#claim

## Build it with a human

Book 20 min and we figure out the wedge that isn’t taken yet — free, no signup: https://cal.com/sattyamjjain