AI Daily Digest · 2026-07-29
🔬 New AI Craft
1. T3MP3ST — Multi-Agent Red Teaming Platform for LLM Security (5263⭐)
T3MP3ST is an autonomous red teaming platform that uses multiple AI agents to automatically discover security vulnerabilities in LLM applications. One agent attacks, another analyzes defenses, a third coordinates strategy — forming a closed-loop offensive security pipeline. Unlike static code analysis tools used for AI Code Review, this validates system-level security, catching prompt injection and data leakage risks.
https://github.com/elder-plinius/T3MP3ST
2. Finn-loop — 3-Skill AI Software Factory for Claude Code
Finn-loop decomposes AI software development into three independent agent skills: spec (requirements), build (coding), and review (code review), with humans merging at the end. Unlike single-agent end-to-end approaches, this uses three focused skills in a relay, enabling per-stage quality gates in team workflows.
https://github.com/finna/Finn-loop
3. Opus 5 Scores 24% on SlopCodeBench — Long-Horizon Coding Still Needs Human Steering
SlopCodeBench (Mar 2026) is a long-horizon coding benchmark where models don't see the full spec upfront — they must evolve codebases across multiple checkpoints. Opus 5 hit 24% (vs Opus 4.6's 17% and GPT-5.4's 11%), but wrote 5× the functions of Opus 4.8. Key signal: even frontier models can't yet do real software engineering without human oversight.
https://github.com/humanlayer/advanced-context-engineering-for-coding-agents/blob/main/benchmarking-opus-5-on-slop-code-bench.md
🛠️ Tools & Tips
1. OptMem — 426 Tokens for Permanent AI Agent Memory
A single 426-token prompt and script gives any AI agent persistent, cross-session memory — plug and play.
https://github.com/VictorTaelin/OptMem
2. no-ai-slop — Strip AI Writing Patterns from Any Text
Detects and removes 20+ AI writing patterns, preserving the author's natural voice. Great for cleaning up AI-generated drafts.
https://github.com/petergyang/no-ai-slop
3. open-connector — Universal Auth Gateway for 1000+ SaaS to AI Agents
Open-source auth gateway connecting 1000+ SaaS providers to AI agents through SDK, CLI, MCP, HTTP, and OpenAPI. Solves the authentication headache in agent toolchains.
https://github.com/oomol-lab/open-connector
4. penecho — AI Canvas Beyond the Chat Box
A shared canvas for handwriting, equations, diagrams, and spatial reasoning — extends AI interaction from "chat" to "whiteboard."
https://github.com/penecho/penecho
⭐ Open Source Highlights
1. Kimi-K3 by MoonshotAI — Open Frontier Intelligence (1426⭐ in 2 days)
Moonshot AI's next-gen open-source LLM, positioned as "Open Frontier Intelligence."
https://github.com/MoonshotAI/Kimi-K3
2. synthetic-sciences/openscience — Open-Source AI Research Workbench
Local-first, model-agnostic AI workbench for scientific research. Multiple open-source variants released on the same day (~4800⭐ combined).
https://github.com/synthetic-sciences/openscience
3. slvDev/esp32-ai — Run AI on ESP32 Microcontrollers
Enables AI inference on ESP32 MCUs (1751⭐), minimal footprint for edge AI deployment.
https://github.com/slvDev/esp32-ai
📰 Industry News
1. Anthropic CEO Dario Amodei publishes company's position on open-weights models, weighing safety vs benefits of open-source AI
2. US judge rejects Google's attempt to use DMCA takedowns to prevent web scraping of its services
3. Netflix employee terminated after sharing personal details in company retreat trust exercise, sparking privacy debate
🚀 Major Releases
1. Microsoft launches MAI-Cyber-1-Flash, an AI cybersecurity model integrated into the MDASH security platform
2. Moonshot AI open-sources Kimi-K3, its next-generation frontier intelligence model
