{"slug": "herdr-notes", "title": "herdr notes", "summary": "Herdr, a terminal multiplexer for AI agents, enables a lead agent to coordinate multiple coding agents across panes via a CLI. The tool requires installing integrations for agents like Claude, Codex, Grok, and Kimi, and uses a skill to teach the lead the CLI, preventing it from inventing MCP or ignoring other panes. The workflow includes starting agents with skip-permissions flags, listing agents, and prompting them with specific commands, with Git operations handled by the lead.", "body_md": "The whole product is three steps:\n\n1. Open Herdr in the repo.\n2. Open the agents you want in other panes.\n3. Tell the lead to talk to them with the Herdr CLI.\n\nThe skill is the how-to. The prompt is the job.\n\nOne integration per agent:\n\n```\nherdr integration install claude\nherdr integration install codex\nherdr integration install grok\nherdr integration install kimi\nherdr integration install antigravity-cli   # agy\n```\n\nTeach the lead the CLI:\n\n```\nnpx skills add herdrdev/herdr --skill herdr -g\n```\n\nWithout the skill, the lead often invents MCP or ignores the other panes.\n\n```\ncd <repo>\nherdr\n```\n\nOpen a window for each agent:\n\n- New tab: click `+` , or prefix then`c`\n- Split: right-click the pane frame, or prefix then `v`\n\nStart each helper in its own pane, optionally with that CLI's skip-permissions flag:\n\n| Agent | Flag | \n|---|---|\n| claude, agy | `--dangerously-skip-permissions` | \n| codex, grok, kimi | `--yolo` | \n\nThen:\n\n```\nherdr agent list\n```\n\nUse those names exactly. Do not open every agent for every task. Two helpers plus a lead is enough.\n\nThe lead must be running *inside* Herdr (`HERDR_ENV=1`). Copy names from the list. Do not put the lead on the call list.\n\nPaste into the lead. Swap roles and the task. Need `gh` logged in (`gh auth status`) for the PR step.\n\n```\nYou are the lead. Use the herdr CLI to talk to the other agents\nin this session. Run `herdr agent list` if you need names.\nDo not invent MCP, subagents, or tmux.\n\ngrok — <role>\ncodex — <role>\nOnly <writer> edits files.\n\nTask: <one sentence>\n\nGit:\n- New branch off main. Commit when the work is done.\n- Open a PR. Do not merge. Do not push to main.\n\nStop when the PR URL exists. Paste it.\n```\n\nUse this when the lead does the work itself, cannot see the other panes, or sits on `agent_blocked`.\n\n```\nStop. You are inside Herdr. Use only the herdr CLI.\n\n1. herdr agent list\n2. Talk only to those names:\n   herdr agent prompt <name> \"<job>\" --wait --timeout 300000\n   herdr agent read <name> --source recent-unwrapped --lines 200\n3. If the command returns agent_blocked: do not retry.\n   Tell me the pane and wait.\n4. If HERDR_ENV is not 1, say so and stop.\n\nGit still: branch, commit, PR. No merge.\nThen continue the task.\n```\n\n| What you want | What to do | \n|---|---|\n| Leave, keep agents running | `Ctrl+Space` , then`d` (Omarchy). Stock Herdr: prefix then`q` . Closing the window is the same. | \n| Kill the session | `herdr server stop` | \n| `prefix+x` closes a pane.`prefix+k` closes a tab. Neither stops the server. |  | \n| Reattach later | `herdr` or`Super+Ctrl+Return` |", "url": "https://wpnews.pro/news/herdr-notes", "canonical_source": "https://gist.github.com/bradflaugher/99f364cee8ee6ad2f65360839eb3961b", "published_at": "2026-09-08 18:31:15+00:00", "updated_at": "2026-09-09 12:41:33.870416+00:00", "lang": "en", "topics": ["developer-tools", "ai-agents", "ai-tools"], "entities": ["Herdr", "Claude", "Codex", "Grok", "Kimi", "Antigravity CLI"], "alternates": {"html": "https://wpnews.pro/news/herdr-notes", "markdown": "https://wpnews.pro/news/herdr-notes.md", "text": "https://wpnews.pro/news/herdr-notes.txt", "jsonld": "https://wpnews.pro/news/herdr-notes.jsonld"}}