- unwind ai
- Posts
- Multi-Agent Kanban board in Hermes Agent
Multi-Agent Kanban board in Hermes Agent
+ Make Claude Code talk less and ship more
Shubham Saboo & Gargi Gupta
May 04, 2026
Today’s top AI Highlights:
& so much more!
Read time: 3 mins
AI Tutorial
Your agent has a 200k token context window.
The 400 tokens of instructions it actually needs are buried under tool definitions, reference docs, and brand guides it never asked for. So it ignores them.
This is the most common reason agents fail in production. It's not a model problem or a framework problem.
In this blog, you'll learn the anatomy of Agent Skills: why the first two lines of SKILL.md are the most important writing you'll do, and how the LLM itself routes queries to the right skill without embeddings or retrieval layers.
Read on to learn the five parts that make skills work, then pick one workflow you do every week and ship your first skill today.
We share hands-on tutorials like this every week, designed to help you stay ahead in the world of AI. If you're serious about leveling up your AI skills and staying ahead of the curve, subscribe now and be the first to access our latest tutorials.
Latest Developments
Why use many token when few token do trick?
Caveman is a Claude Code skill and plugin that makes Claude talk like a caveman. It cuts roughly 75% of output tokens while keeping every bit of technical accuracy intact.
Claude’s responses and fillers like "Sure, I'd be happy to help you with that" cost tokens and add zero value. With Caveman active, Claude drops articles, kills pleasantries, and eliminates hedging, but leaves code blocks, technical terms, and error messages completely untouched. The result is faster responses, lower API costs, and honestly, funnier code reviews.
Key Highlights:
75% token reduction – Strips filler words, hedging, and pleasantries from Claude's responses while preserving all technical content, code blocks, and exact error messages.
One-line install – Run
claude install-skill JuliusBrussee/cavemanAnd you're set.Toggle on demand – Activate with
/cavemanor "caveman mode" plugin and switch back to normal Claude anytime with "stop caveman" or "normal mode."
Mistral just shipped remote agents in Vibe - async coding sessions that run in the cloud, execute in parallel, and ping you when they're done.
Powered by Mistral Medium 3.5, a new 128B dense model built for long-horizon agentic tasks. The model merges instruction-following, reasoning, and coding into a single set of weights, scores 77.6% on SWE-Bench Verified, and is now the default across both Vibe CLI and Le Chat.
ICYMI: Vibe's CLI is an open-source terminal-native coding agent, working inside your codebase with slash-command skills, custom subagents, and MCP integrations.
Key Highlights:
CLI or Le Chat — Spawn remote coding agents from either surface.
Teleport local sessions to the cloud — Move an ongoing CLI session to a remote sandbox mid-task, with history, state, and approvals carrying across. When done, the agent opens a PR on GitHub and notifies you.
Mistral Medium 3.5 — Open-weights released under modified MIT, self-hostable on as few as four GPUs, with configurable reasoning effort.
Work mode in Le Chat — A new agentic mode that handles multi-step tasks like inbox triage, research synthesis, and cross-tool workflows, with connectors on by default and explicit approval for sensitive actions.
Browser Use is a tens-of-thousands-line browser framework. Now, the team turned around and shipped a 592-line alternative that lets the LLM be the framework.
Browser Harness is a new open-source project that gives an LLM a raw CDP websocket connection to Chrome, a small set of starter helpers, and a skill file - that’s it!
The agent sees exactly how its tools work at the protocol level, writes new ones when it needs them, and self-corrects when something breaks.
In one interesting example shared by the team, the agent needed to upload a file and it didn’t have the tool, so it wrote one itself and moved on.
And now, after Browser Harness blew up, the team just packaged it into Browser Use Desktop, an open-source desktop app that lets Claude Code, Codex, or any coding agent control your browser, without trying to replace your Chrome.
You can even trigger these agents by texting yourself @BU on WhatsApp.
Quick Bites
Multi-agent task coordinator Kanban in Hermes Agent
The latest version of Hermes Agent ships a Kanban multi-agent system where agents, each with their own tools, skills, and personality profiles, claim tasks off a board, fan out work through linked dependencies, and pass files via shared workspaces or git worktrees. There's a live dashboard, per-task comment threads that both humans and agents write into, heartbeat monitoring, and the whole thing is SQLite-backed so it survives crashes and reboots.
20 Claude Code tips you probably missed
This one’s a nice recap or fresher on the best hotkeys and commands to use Claude Code. It's one of those lists where you go in thinking you know most of it and come out humbled. We definitely didn’t know about 11 and 13 in the list. Worth bookmarking!
Agents can now create account, buy domain, and deploy
Cloudflare and Stripe just teamed up so coding agents can go from zero to deployed autonomously. This includes creating a Cloudflare account, buying a domain, starting a paid subscription, and shipping to production, all without any human intervention anywhere. Stripe acts as the identity provider and handles payment tokenization (with a $100/month spending cap per provider, so your agent doesn't go on a domain shopping spree), while Cloudflare auto-provisions accounts on the fly. Just install Stripe CLI + Stripe Projects plugin, and you can start! The team will also release an open protocol soon.
Tools of the Trade
Deepclaude - A proxy that lets you swap Claude Code's backend for DeepSeek V4 Pro, OpenRouter, or any Anthropic-compatible API. You keep the same Claude Code agent loop and UX but at roughly 17x lower cost.
Agent-desktop - A Rust-based CLI that gives AI agents native desktop control by reading OS accessibility trees and outputting structured JSON with stable element references. Think browser automation, but for any native app on your machine.
Heard - A voice companion that reads your AI coding agent's terminal replies aloud so you don't have to keep watching the screen. Wispr handles what you say to your agent; Heard handles what it says back. It monitors agent output and uses TTS to keep you in the loop while you do other things.
Awesome LLM Apps - A curated collection of LLM apps with RAG, AI Agents, multi-agent teams, MCP, voice agents, and more. The apps use models from OpenAI, Anthropic, Google, and open-source models like DeepSeek, Qwen, and Llama that you can run locally on your computer.
(Now accepting GitHub sponsorships)
That’s all for today! See you tomorrow with more such AI-filled content.
Don’t forget to share this newsletter on your social channels and tag Unwind AI to support us!
PS: We curate this AI newsletter every day for FREE, your support is what keeps us going. If you find value in what you read, share it with at least one, two (or 20) of your friends 😉
Reply