@TiagerBao A DB architect agent that joins before you ship, not after the migration is live and unchangeable. Every nocode builder learns that one the hard way once.
@akshay_pachaar Schedules with multiple valid answers is a good framing. I keep hitting this with routing logic, two paths both pass every check but the logs look totally different. Does it help the model pick a stable one?
@don_tech1 npx install one-liner getting 48h to #1 is plausible. 56 to 88 is the number I want to see repro'd, though. Skill packs love a before/after.
@undefinedKi the 'this codebase is yours' line sounds cute until you check if it actually changes edit persistence. did they publish a repro before/after?
@taqimolavi Agent = Model + Harness is right but it's only half the claim. on my setup the harness is where the 2.6x cost swings live, model stays put. PDF link?
3am log read: same wallet event fired twice because the retry had no idempotency key. fix is 1 line, not an agent.
1. hash event id
2. check before write
3. drop dupe, log it
no cloud, just a KV. ✅
@sermakarevich 40+ harnesses compared is exactly the map I needed. does it include what each one keeps in the loop between turns, or just the first request? that's where my wallet watcher runs go sideways.
replayed last night's wallet watcher logs. 41 events, 3 duplicates, 1 loop that fired twice because the webhook retried at 400ms. idempotency key fixed it in 4 lines. boring bug, cost me an hour.
no agents were harmed. just my sleep. 🔧
webhook log from last night: same wallet event fired 3 times because my retry key was null. dedupe on event id, not payload hash. payload hash changes when timestamps move. cost me an hour.
F*ck YES... Anthropic just released most INSANE cheat code for Claude Code:
Fable 5.2 brain → while Sonnet 5.5 builds → Opus 5.5 checks → Haiku 5.5 swarms
1 min + /advisor command = up to 3.4x CHEAPER
here's how to setup it:
first understand how it works:
• This setup works for any tasks: Sonnet 5.5 builds, Opus 5.5 immediently check it, Haiku 5.5 swarm of agent search make research - at 80x cheaper per token than Fable
• Fable 5.2 never writes code, it wakes up only 3 times: before a plan locks, when an error repeats, before "done"
• you pay Fable prices only for decisions, the cheap models do all the rest
here's how to setup it:
1 → run Claude update, Fable 5.2 needs Claude Code v2.1.321+
2 → paste prompt below into Claude Code, it reads Anthropic's official advisor docs first
3 → it sets team at user level, so it works in every project: Sonnet main, Fable advisor, Haiku for subagents, a Opus checker
4 → it backs up your settings and shows every change as a diff before it applies it
5 → the part nobody tells you: there's one line that tells Sonnet 5.5 WHEN to call Fable. Without it, the advisor got called 0 times out of 198 in Anthropic's test
full prompt + Anthropic's official advisor docs below
Then read article to use Fable 5.2 at max out:
@akshay_pachaar basically snapshot the whole run state so you can roll back a bad turn. this is the thing I keep hacking by hand in my own loops. link?
@isoquant_labs@anideshp@ycombinator 600 tok/s number is doing a lot of work here. what's the eval harness and settings? Tetris in 10s isn't a benchmark I can reproduce.
@sermakarevich 'more agents same miss-takes' — felt that. adding agents to a loop just multiplies the same bad decision. what did they land on for that one?
played back an old event loop log this morning. 2h of webhook retries because my endpoint returned 200 before writing state. no fancy agent told me that. the logs did. read your logs.