AI/ML●●●Banger
Retrace fork a failed AI agents run, replay it, prove the fix
Fork from failed agent runs and prove fixes before shipping—LangSmith doesn't do this.
Solve My ProblemShip It
Yashwanthbogam
3011d ago
Back to browse
GitHub Repository
Evidence home for the verdict-layer approach: pre-registered, reproducible trust evaluation of AI-generated code changes - silent wrong ships 17/50 → 0/50.
1 starsPython
I pre-registered the rules,then measured AI agents shipping wrong fixes
by kolesnikov-arch·Jul 6, 2026·1 point·1 comment
Post Description
A controlled comparison of ungated vs gated AI patching on SWE-bench Lite. Full methodology, results, and reproducible scripts in the repo. Breakthrough for flash models.
Similar Projects
Developer Tools●●Solid
Agents ship diffs. Visr keeps trajectories
Trajectory tracking beats diffs for agent memory, but terminal recording isn't new.
Big BrainShip ItNiche Gem
sourishkrout
401mo ago
AI/ML●●Solid
I made a compiler/VM for untrusted scripts
Wraps existing libraries so AI can't touch the filesystem unless you explicitly allow it.
Big BrainNiche Gem
hoansdz
201mo ago
Developer Tools●●●Banger
LawClaw – Constitutional governance for AI agents (MIT)
Constitutional governance for AI agents beats prompt injection with pre-execution enforcement.
Big BrainSolve My ProblemZero to One
nghiahsgs
214mo ago
Developer Tools●●Solid
Agent from Scratch – Bootstrap an agent from a copy-paste, no framework
Agent self-bootstraps with no frameworks, but novelty wears fast without real problem solved.
Big BrainWizardryRabbit Hole
brotchie
304mo ago
Infrastructure●●Solid
Argus – Self-hosted AI that monitors infra and explains what's wrong
ReAct agent with 18 tools investigates incidents before you open Datadog dashboards.
Solve My ProblemShip It
PreciousH
103mo ago