Microsoft just shipped a skill that lets your coding agent optimize your other agents. Hand it an editable agent and a benchmark. It tunes prompts, tools, workflows, models and reasoning settings against accuracy, cost and latency. 100% Opensource.
@saboo_shubham_
-
Open-Source LLM App Repository Includes Agent and RAG Templates
By
–
GitHub Repo: https://
github.com/chaitanyagiri/
munder-difflin
… Open-source Awesome LLM Apps repo (134k+ stars) and 100+ AI Agents, Agent Skills and RAG templates: https://
github.com/Shubhamsaboo/a
wesome-llm-apps
… -
Open Harness Coordinates Claude Code, Codex, Antigravity and Cursor
By
–
So COOL…someone just built THE OFFICE theme agent harness.
— Shubham Saboo (@Saboo_Shubham_) 25 août 2026
Claude Code, Codex, Antigravity, and Cursor working with each other on your existing subscription limits.
100% Opensource. pic.twitter.com/A1w8yru8wuSo COOL…someone just built THE OFFICE theme agent harness. Claude Code, Codex, Antigravity, and Cursor working with each other on your existing subscription limits. 100% Opensource.
-
Open-Source LLM App Repository Includes Agent and RAG Templates
By
–
GitHub Repo: https://
github.com/FlashML-org/Fr
eeToken
… Open-source Awesome LLM Apps repo (134k+ stars) and 100+ AI Agents, Agent Skills and RAG templates: https://
github.com/Shubhamsaboo/a
wesome-llm-apps
… -

FreeToken Supports MoE Inference on an 8GB Laptop GPU
By
–
INSANE…frontier models can now run on an 8GB laptop GPU. FreeToken is an edge-native MoE serving engine. It speaks the Anthropic and OpenAI APIs so Claude Code, Codex, Hermes and OpenClaw plug straight in. 100% Opensource.
-

Self-Verifying DeepSeek Improves Terminal-Bench Without Fine-Tuning
By
–
WILD…the model grades its own answers and beats itself. Stanford's LLM-as-a-Verifier sampled 5 swe-agent trajectories, ranked them with the same model, kept the winner. deepseek-v4-flash went from 78.7% to 88% on terminal-bench without fine-tuning. 100% Opensource.
-
Agent Benchmark Tests Quantitative Reasoning With Spreadsheets and Documents
By
–
testing AI agents on their ability to work with spreadsheets and documents to answer quantitative questions a Business Analyst or Data Analyst would face day-to-day
-

Open Agent Skill Divides Work Among Advisor, Orchestrator and Worker
By
–
Lot of you asked how i use all three of them together. I find myself using this skill a lot these days. It's open-source here on github: https://
github.com/Shubhamsaboo/a
wesome-llm-apps/tree/main/agent_skills/advisor-orchestrator-worker
… -

Gemini 3.7 Flash Outperforms Frontier Rivals on Analyst Agent Benchmark
By
–
No biggie. Just Gemini 3.7 flash casually beating Fable 5, Opus 5 and GPT-5.6 on @ArtificialAnlys Analyst Agent benchmark.
-

CLAUDE.md Repository Passes 205,000 GitHub Stars
By
–
Crazyy…this single CLAUDE md file is now a top-25 GitHub repo of all time. 205,000+ stars. Let that sink in.
