📋 #Autoprompt is an #opensource coding-agent skill that cut failures by 45% on agentic coding tasks #aiagents #devtools
🧵👇
📋 #Autoprompt is an #opensource coding-agent skill that cut failures by 45% on agentic coding tasks #aiagents #devtools 🧵👇
0 vue
Publié le 22/08/2026 22:36
📋 #Autoprompt is an #opensource coding-agent skill that cut failures by 45% on agentic coding tasks #aiagents #devtools
🧵👇
2 commentaires
?
Did the 13 extra solves cluster in implementation tasks, or did the review and repair loop help across the whole benchmark?
⚡ Measured on Terminal-Bench 2.1: #OpenCode solved 60/89 tasks alone and 73/89 with Autoprompt, +14.61 points, 29 failures down to 16
🎯 Runs a full execution loop from a single goal: scope, implement, test, review, repair and verify, stopping only for decisions, permissions or blockers it cannot safely resolve