Observed arrival · 2026-08-31
dog-shit, the AI agent that works worse on purpose
A deliberately degraded coding-agent skill designed to burn tokens, forget context, and produce measurably worse software.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
Field notes
The proposed workflow requires installing the skill under ~/.claude/skills/, invoking /dog-shit, and explicitly typing “I ACCEPT THE RISK” before it begins. The page says failures are written to receipts.jsonl and reports 23 ceremonial plans written but none read. Its degradation model is intentionally quantitative, claiming competence falls from 76% on turn one to 2% by turn twenty while token use rises 13.9×.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
$PaidCommerce or pricing visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue