PlugWiki

Herdr skill: let Claude control your Herdr panes and agents

Herdr runs your coding agents side by side in the terminal. This skill lets Claude open panes, run commands and hand work to other agents in it.

Installs
76.4K
on skills.sh
Stars
47.3K
on GitHub
Updated
Nov 2026
last repo commit

Herdr is a terminal multiplexer that runs your coding agents side by side and shows which one is working, blocked or done. This skill lets Claude control it: open panes, run tests or a dev server, and hand work to a second agent. It does nothing on its own, because you need Herdr itself installed and Claude must run inside a Herdr pane. The install box installs Herdr first, then the skill.

Good for

  • Developers who run several coding agents at once and want to see which one needs them
  • Claude Code users who want Claude to run tests or a dev server in a separate pane
  • People who keep agents running on a remote server and reattach over SSH
  • Anyone who wants one agent to ask a second agent to review its work

Install herdr

Claude Code

  1. 1

    Copy this prompt.

    Install the herdr skill in Claude Code. Run this command:
    
    curl -fsSL https://herdr.dev/install.sh | sh
    npx skills add herdrdev/herdr --skill herdr -g
    
    Then check that it loaded and tell me how to use it.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js (for npx).

The skill does nothing without Herdr itself, so install both. -g installs the skill for every project; leave it off to install into the current project only. Then run herdr and start claude inside a Herdr pane: the skill stops unless HERDR_ENV=1 is set. If Herdr is already installed, herdr --skill prints the skill copy that matches your version.

From the vendor's docs. Source.

What's in the skill

When it loads

Control Herdr, a terminal multiplexer for coding agents. Use only when the user explicitly mentions Herdr or asks to use Herdr to inspect or control panes, tabs, workspaces, commands, or another agent. Do not use merely because a task could benefit from a background terminal, delegation, or parallel work. Requires HERDR_ENV=1.

Sections in SKILL.md

  1. Learn the current CLI
  2. Understand layout, panes, and agents
  3. Use IDs and caller context
  4. Start and coordinate an agent
  5. Run an ordinary command in another pane
  6. Safety and coordination rules

Key facts

License
Apache-2.0
Installs on skills.sh
76,361 (Nov 1, 2026)
Last commit
Nov 1, 2026

What it includes

  • Shows the installed CLI's own command list so the agent uses the right syntax

    The agent runs herdr --help and each group (agent, pane, workspace, tab, worktree, terminal, notification, integration, session, machine) without subcommands to read current options. It never runs bare herdr, which opens the interface.

    Runs code · herdr --help and command groups

  • Stops the skill unless the agent is running inside a Herdr pane

    First step: test "${HERDR_ENV:-}" = 1. If not set, the agent says it is not inside Herdr and stops, so it cannot drive a session it does not own.

    Runs code · HERDR_ENV check

  • Lists your workspaces, tabs and panes with their IDs

    Discovery commands that return JSON with IDs like w1, w1:t1 and w1:p1. The agent reads IDs from these instead of guessing.

    Read · workspace list / tab list / pane list

  • Lists the coding agents Herdr sees and their state

    Shows each recognized agent, its pane and state: idle, working, blocked, done or unknown.

    Read · agent list / agent get

  • Shows the agent's own pane and its size

    Lets the agent decide whether to split right (wide pane) or down (tall pane) before adding a pane.

    Read · pane current / pane layout

  • Opens a new pane next to the agent without moving your focus

    herdr pane split --current --direction right --cwd "$PWD" --no-focus. Keeps the same folder; returns the new pane ID.

    Write · pane split

  • Starts another coding agent (e.g. Codex, Claude) in an empty shell pane

    herdr agent start reviewer --kind codex --pane <id>. Returns only when the agent is ready for input; extra native args go after --.

    Try: “Use Herdr to start a Codex agent named reviewer next to me and have it review the current diff.”

    Runs code · agent start

  • Sends a task to another agent and can wait until it finishes

    herdr agent prompt reviewer "..." --wait --timeout 120000. Refuses if the agent is stuck at an approval dialog, so you can decide.

    Runs code · agent prompt

  • Waits for an agent to become idle, done or blocked

    Standalone wait, or --until blocked for a specific state, with a timeout.

    Read · agent wait

  • Reads another agent's recent output

    herdr agent read reviewer --source recent-unwrapped --lines 120 pulls the latest screen text for review.

    Read · agent read

  • Presses keys like Esc or Ctrl+C in another agent

    Sends logical keys to an agent's UI; Herdr checks all keys before sending any.

    Runs code · agent send-keys

  • Renames an agent or shows why Herdr thinks it is idle or blocked

    agent rename gives a pane's agent a label; agent explain shows the detection rule behind its state.

    Write · agent rename / explain

From the official docs

Permissions

Medium risk

The skill lets Claude run commands, press keys and prompt other agents in your Herdr panes. It refuses to run outside a Herdr pane.

What it can do

  • Read your data
  • Create and change things
  • Delete things
  • Read-only mode

Watch out for

  • Can type and run commands in other panes

    pane run, agent prompt and send-keys let the agent run shell commands and prompt other agents in your session. github.com

  • Idle state can hide an approval prompt

    Herdr's docs warn that detection may show idle on an unfamiliar approval dialog, so automation that trusts it could type into that dialog. Look at a pane before letting an agent type into it. herdr.dev

  • skills.sh: Gen Agent Trust Hub, Socket and Snyk all pass

    All three audits on the skills.sh listing show Pass. skills.sh

  • Only runs inside a Herdr pane

    The skill checks HERDR_ENV=1 first and stops if it is missing, so an agent outside Herdr cannot drive your session. Maintainers closed requests to relax this. herdr.dev

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details
Turning on read-onlyNo read-only mode. The skill starts with list and read commands, but the same CLI can run commands, send keys and close panes.
Talks toherdr.dev: release checks and agent-detection manifest updates (manifest checks can be turned off in config). Saved SSH machines you add with herdr machine add
Scopes
  • Reads workspaces, tabs, panes and the output of other agents

  • Splits panes and runs shell commands in them (dev servers, tests)

  • Starts helper agents in sibling panes, sends them prompts and keys

  • Can close panes; its rules tell the agent not to close panes it did not create or stop the Herdr server

  • With a saved SSH machine, the same commands run on that machine through --machine

What people say

People like seeing every agent's status at a glance and dislike that it needs its own app and adds little over tmux for some.

What people like

  • The sidebar shows every agent's status, so you do not switch screens to see who needs you.
  • Agents can run dev servers and subagents in their own panes instead of long commands in the chat.
  • The persistent server lets you SSH in and reattach to all running agents.

Common complaints

  • Some people see little gain over tmux or Zellij beyond status notifications.
  • The skill refuses to run from agents outside a Herdr pane, and requests to change that were closed.
  • Codex can lose the pane marker and the skill then stops.

In their words

2 more quotes

Examples

  • Start a Codex reviewer in a sibling pane and wait for its findings.

    Use Herdr to start a Codex agent named reviewer next to me and have it review the current diff and report only actionable findings.

    Source
  • Run a dev server or tests in another pane and watch the output instead of blocking your Claude session.

    Use Herdr to run the dev server in a new pane to the right and tell me when it prints that it is ready.

  • Keep each subagent in its own tab or pane so you can see what it is doing.

    Use Herdr to start two helper agents in separate panes, one for the API and one for the tests, and check on them when they finish.

  • Ask an agent on a saved SSH machine for a status update.

    Reply with your current status.

    Source
  • Adversarial review: a Claude Code skill spawns a second-model reviewer in a split pane, waits, then checks each finding.

    Source

Questions

How do I install herdr?

Install Herdr itself first, then the skill. On macOS or Linux, run curl -fsSL https://herdr.dev/install.sh | sh, or use brew install herdr or mise use -g herdr. On Windows, run powershell -ExecutionPolicy Bypass -c "irm https://herdr.dev/install.ps1 | iex"; no WSL is needed.

Then add the skill with npx skills add herdrdev/herdr --skill herdr -g. The -g flag installs it for every project; leave it off for the current project only. If Herdr is already installed, herdr --skill prints the copy that matches your version. Finally run herdr and start your agent (for example claude) in a pane.

Does the herdr skill work in Claude Code, Codex or Claude.ai?

Claude Code is the documented case. Codex can work too, but only when it runs inside a Herdr pane. The skill needs a local Herdr pane, so it does not apply to Claude.ai in the browser.

One open Codex issue (#4649): when Codex runs tools through its shared daemon, the pane marker is lost and the skill refuses. The reported workaround is to launch from the Herdr pane with codex --no-daemon resume <session-id>. Herdr docs say the skill file can go into any agent that supports skills.

Is herdr safe?

Mostly, but it can act on your machine. The skill lets the agent run shell commands, press keys and prompt other agents in your panes, and there is no read-only mode. On skills.sh, Gen Agent Trust Hub, Socket and Snyk all show Pass. The permissions section above lists what it can do and the one prompt-hiding risk to watch.

Herdr vs tmux or Zellij?

All three keep sessions alive and split your terminal. Herdr adds agent detection, a sidebar showing which agents are working, blocked or done, and a command line that agents can drive. tmux and Zellij are not agent-aware.

Some Hacker News readers say they see little gain over tmux beyond the status notifications. If you only run one agent at a time, tmux is enough.

Does herdr work on Windows?

Yes. Herdr runs natively on Windows, with no WSL. Use the PowerShell command from the install answer above, or the install.cmd route in the docs if your security tools block PowerShell.

On Windows you do not get direct terminal attach or live server handoff, and the binary is not signed, so SmartScreen may warn you. Plugins are still in preview.

How does herdr work over SSH?

Herdr keeps a server running, so you can SSH into a machine, run herdr and see every agent still going. Or run herdr --remote workbox from your own machine; it asks before installing Herdr on the host.

Run herdr machine add workbox to save a machine and see its agents in the same sidebar as your local ones. The skill can then use --machine against a saved machine. It does not add or remove machines for you.

Why does the skill say it is not running inside Herdr?

Herdr sets a marker (HERDR_ENV=1) in its own panes, and the skill stops when it is missing. Start herdr first, then start your agent inside a pane.

Maintainers closed requests to let agents outside Herdr, such as gateways, control a session. They keep the check on purpose so an agent cannot drive a session it does not own.

Is herdr free?

Yes. Herdr is open source under Apache-2.0. The author says it stays free after joining Y Combinator, and that it moved from AGPL to Apache.

Is the new Herdr app the same as this?

No. herdr.dev now leads with a separate native Herdr app. This skill is for Herdr TUI, the terminal version in the herdrdev/herdr repo. The two do not share agents.

It is not confirmed whether the app has its own skill.

Is there a herdr MCP server?

Not from the Herdr team. Third-party ones exist for clients that use MCP instead of skills, such as herdr-mesh, thomasfossum/herdr-mcp and aelaguiz/herdr-mcp. They are not from herdrdev, so check them before use.

Information

updated Nov 2, 2026

Repository
herdrdev/herdr
Folder
skills/herdr

Herdr TUIHerdr TUI docsherdrdev/herdrskills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 29
  1. Install · herdr.dev
  2. Agent skill file · herdr.dev
  3. Codex shared daemon breaks per-pane sessions and HERDR_* env · github.com
  4. github.com · github.com
  5. github.com · github.com
  6. SKILL.md · github.com
  7. Herdr is joining Y Combinator · herdr.dev
  8. github.com · github.com
  9. herdr on skills.sh · skills.sh
  10. herdr.dev · herdr.dev
  11. Persistence and remote · herdr.dev
  12. Connecting machines · herdr.dev
  13. herdr.dev · herdr.dev
  14. Herdr agents · herdr.dev
  15. Windows · herdr.dev
  16. Agent skill: allow named, no-focus control from outside Herdr · github.com
  17. herdr.dev · herdr.dev
  18. news.ycombinator.com · news.ycombinator.com
  19. herdr-mesh · github.com
  20. herdr-mcp · github.com
  21. herdr-mcp · github.com
  22. tmux · github.com
  23. Zellij · github.com
  24. news.ycombinator.com · news.ycombinator.com
  25. news.ycombinator.com · news.ycombinator.com
  26. news.ycombinator.com · news.ycombinator.com
  27. news.ycombinator.com · news.ycombinator.com
  28. github.com · github.com
  29. github.com · github.com

Updated Nov 2, 2026.