OSSQuesmaOrg/BinaryAudit — AI Binary Backdoor Detection BenchmarkGitHub·high signalXBlueskyLinkedInCopy link33 tasks, 17 models. Claude Opus 4.6 leads at 49%. Hidden backdoors in real servers. Ghidra + AI agent workflow.SourceSource pageGitHub↳ Follow the threadStack layer / ContrastCROSS-CATEGORY: Four Independent Projects in 48 Hours Built the Portability Layer That Moves Lock-In Off the Agent VendorEngrim, Kit, Yurei and Crew (via Hacker News Show HN and Product Hunt)Stack layer / ContrastSpeakeasy launched Kit, an MIT Rust coding-agent runtime that gives the model one compose tool and reports half of Claude Code's input tokensGitHub / SpeakeasyStack layer / ContrastExperiential Labs Ships an Open-Source, Zero-Markup OpenRouter That Distills a Model You Own From Your Own TrafficGitHub API (corroborated by Product Hunt and ycombinator.com company page)Stack layer / Follow-up threadFrontierHarness Eval: same model, nine harnesses, 17.5x cost spread and a 16-point pass-rate spreadGitHubStack layerPenelopa.ai mines real Codex and Claude Code session logs and turns repeated workflows into skillsGitHubPolicy dependency / Stack layerwebmcp-stack Generates Agent Tool Surfaces From an OpenAPI Spec So the Safety Decisions Survive Regenerationwebmcp-stack, via Hacker News Show HN (single source)Stack layerMagnitude is a local-inference server that plugs into eight existing coding agents rather than replacing themGitHub TrendingThreat patternLiteLLM v1.100.0 deletes prompt_token_calculator and signs every image with cosignGitHub