6 Comments
User's avatar
Rohan's avatar

love that the fantasy agent caught a starter on bye. did you add the review gate after an agent faked a done?

Lily Luo's avatar

no, I added the review gate when the agent suggested a bench player of the same position, when it was RB/WR/TE slot. I needed it to do more research and justify who it was replacing based on my team, my opponent (if i needed safe floor or ceiling), who’s on the waiver, etc. so that helped me make better decisions!

Rohan's avatar

making it justify who gets benched is smart. does it ever argue you out of a start?

Lily Luo's avatar

yes! to varying success, since FF is like 80% luck lol

Guru's avatar

The idea that rules are enforced in code, so a task can't close until a reviewer on a different model has passed its latest version, is the part I'd steal first. The campaign email QA example makes it very concrete. Question: for marketing ops people at a big company with no terminal access, which piece would you keep if they only had a chat or Project: the task records, the stages, or the second-model review?

Lily Luo's avatar

Thanks Guru! I'd keep the task records. The acceptance criteria is what make the review and validation possible, and they work anywhere, even as a template in your Project instructions.

In the email QA example, that's your checklist: links go where they should, fields have a fallback when blank, the compliance footer and unsubscribe link are there, and the copy follows your voice rules.

Next I'd keep the review, but run it by hand. Paste the draft, the criteria, and the first set of findings into a fresh chat and ask what was missed. A different model is better if your company gives you access to one (or a different model but same family, like Sonnet if you were using Opus, if no other model is provided), but a clean session with no attachment to the work still gets you good results. I think it's worth the extra step. In one weekend, 18 of my 45 reviews sent work back for legitimate changes.

I'd drop the stages. Without code or a script, a stage is just a label and can’t be enforced, so you become the manual reviewer.

If you want a place to start, the Files-First Starter Kit works in the browser: lilyluo.ai/starter-kits/files-first. Would love to hear what you try!