抄作业 / 做成了什么 / Meta Muse

做了什么

偏 Agent 的二手侦察：从照片学品味、每天自调搜索、按成片判断商品，并把结果推到手机，走 Meta Muse MCP 连接器。

场景

未归类

资料完整度

• ✓ 作者与原链接有
• ✓ 做了什么有
• ✓ 来源原文摘录有
• ✓ 截图 / 媒体有
• ✓ 步骤 / 章节有
• — 提示词 / 配置原文来源未提供
• — 结果 / 数字来源未提供
• ✓ 风险提示有
• ✓ 相关条目有

— 表示来源没有给出，本站不补写。

来源原文摘录

AmranHudda/fitscout
Agent-heavy resale scout: learns my taste from photos, tunes its own searches daily, judges listings by their photos, and pushes finds to my phone. Meta Muse MCP connector.

摘自项目 README，照录原文，未翻译。

效果截图

来源里的步骤 / 要点

1. Photos: 50 photos produced 150 garments. My face was matched on-device in 48 of the 50, including 27 group shots.
2. Tuning agent: test-ran 6 searches and rewrote 5 overly specific ones into the words sellers actually use in titles, in 21 seconds for $0.16.
3. First scan: pulled 131 Grailed listings and judged 130 in 103 seconds. It flagged 22 as worth showing and sent the best 6 as notifications.
4. Size check: XXL and M versions of a sweater were marked as the wrong size for "L/XL".
5. Course correction: the top pick was a $43 black Polo quarter-zip at 9.5/10, which is basically one I already own. I said I want new things inspired by my sty……
6. After the correction: the next scan judged 199 listings in 156 seconds. The top picks were a vintage cream Aran wool cardigan ($64), a gray chunky cable-knit……
7. Cost: the whole first night, including building and testing, used $3.12 of OpenAI. A hard daily cap ($2 by default, $5 here) stops the AI calls if it's reached.…
8. Face matching stays on the phone. The cloud model only sees a copy of the photo with a green box around me, so it never has to recognize anyone. My reference……

照录来源原文，未翻译。

提示词

来源没有给出可直接复制的提示词原文。

怎么复制

1. 打开仓库：github.com/AmranHudda/fitscout ↗
2. 读 README，把仓库链接发给你的 Bot，让它按 README 复现
3. 先在小范围跑通，再加 routine

风险与限制

• 定时运行会持续消耗额度：频率从低开始，跑稳再加
• 作者的结果不保证能复现，以原帖为准

用量预估

用金属计算器估一下