What does the Herdr AI skill do?

Control herdr from inside it. Manage workspaces and tabs, split panes, run commands, read output, and wait for state changes — all via CLI commands that talk to the running herdr instance over a local unix socket. Use when running inside herdr (HERDR_ENV=1).

Why use Herdr on TypingMind?

Because you install it once and use it with any model. Herdr is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Herdr in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/ninehills/skills/tree/main/herdr. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Herdr?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Herdr?

As many as you like. As long as a model supports skills, you can use Herdr with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Herdr AI skill free?

It is published on GitHub by ninehills. Check the repository for licensing terms. You only pay your own AI provider for the tokens you use.

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

Herdr by ninehills — AI Skill for Any LLM

herdr — agent skill

you are running inside herdr, a terminal-native agent multiplexer. herdr gives you workspaces, tabs, and panes — each pane is a real terminal with its own shell, agent, server, or log stream — and you can control all of it from the cli.

this means you can:

see what other panes and agents are doing
create tabs for separate subcontexts inside one workspace
split panes and run commands in them
start servers, watch logs, and run tests in sibling panes
wait for specific output before continuing
wait for another agent to finish
spawn more agent instances

the herdr binary is available in your PATH. its workspace, tab, pane, and wait commands talk to the running herdr instance over a local unix socket.

if you need the raw protocol or full api reference, read the socket api docs.

verified version

this skill documents herdr v0.5.8. commands were verified against the actual binary. do not assume newer features (like herdr agent subcommands) exist unless confirmed.

concepts

workspaces are project contexts. each workspace has one or more tabs. unless manually renamed, a workspace's label follows the first tab's root pane — usually the repo name, otherwise the root pane's current folder name.

tabs are subcontexts inside a workspace. each tab has one or more panes.

panes are terminal splits inside a tab. each pane runs its own process — a shell, an agent, a server, anything.

agent status is detected automatically by herdr. the api exposes one public field for it:

agent_status — idle, working, blocked, done, unknown

done means the agent finished, but you have not looked at that finished pane yet.

plain shells still exist as panes, but herdr's sidebar agent section intentionally focuses on detected agents rather than listing every shell.

ids — herdr v0.5.8 uses opaque workspace-scoped pane ids like w651d456f8444f3-2. tab ids look like w651d456f8444f3:2. workspace ids look like w651d456f8444f3. always re-read ids from herdr pane list before using them — do not hardcode or guess.

discover yourself

see what panes exist and which one is focused:

bash
herdr pane list

the focused pane is yours (look for "focused": true). other panes are your neighbors.

list workspaces:

bash
herdr workspace list

do not redirect pane output to files

do not use shell redirection like > /tmp/output.txt or 2>&1 | tee /tmp/log when running commands in a herdr pane via herdr pane run. herdr's value is seeing the live terminal stream — the agent's own output, colours, formatting, and progress indicators. redirecting hides all of that and makes the pane appear frozen.

instead, observe output with:

herdr pane read <id> --source recent --lines N — read what is already in the scrollback
herdr wait output <id> --match "pattern" --timeout N — block until expected output appears
herdr pane read <id> --source visible --ansi — get an ANSI snapshot for TUI feedback loops

pane commands (the only way to interact with panes)

herdr v0.5.8 does not have herdr agent commands. all interaction goes through herdr pane:

herdr pane list — list all panes, find your pane id and neighbors
herdr pane read <id> --source recent --lines N — read what is on the screen
herdr pane send-text <id> "text" — send text without pressing enter
herdr pane send-keys <id> Enter — press a key (Enter, CtrlC, Escape, etc.)
herdr pane run <id> "command" — send text + enter atomically
herdr pane split <id> --direction right|down [--no-focus] [--cwd PATH] — split and create a new pane
herdr pane close <id> — close a pane
herdr pane rename <id> "label"|--clear — assign or clear a pane label
herdr pane get <id> — get detailed info about a pane

sending messages to a pane

herdr pane run <id> "text" sends text + Enter atomically and returns immediately. it does not wait for the receiving side to process the message.

bash
# send a complete message — runs "npm test" in the pane's shell
herdr pane run w651d456f8444f3-3 "npm test"

# send text to an agent in the pane
herdr pane run w651d456f8444f3-3 "review the test coverage in src/api/"

herdr pane send-text <id> "text" sends raw text without pressing enter. use herdr pane send-keys <id> Enter to press enter separately:

bash
# send text without enter
herdr pane send-text w651d456f8444f3-3 "hello"

# press enter
herdr pane send-keys w651d456f8444f3-3 Enter

important: pane run appends \r (carriage return) to the text, equivalent to pressing enter. the command runs in the pane's existing shell process — it does not start a new shell. if the pane is at an agent prompt (e.g. > or the pi input line), the command text will be sent to the agent, not executed as a shell command.

there is no built-in delay or --wait flag on pane run or pane send-text. both return immediately. to pace messages to an agent, use herdr wait agent-status --status idle between sends:

bash
# send a message
herdr pane run w651d456f8444f3-3 "review the test coverage"

# wait until the agent is idle again
herdr wait agent-status w651d456f8444f3-3 --status idle --timeout 120000

# now send the next message
herdr pane run w651d456f8444f3-3 "also check src/utils/"

spawn a new agent

split a new pane, then run the agent in it:

bash
# split to the right without stealing focus
herdr pane split w651d456f8444f3-2 --direction right --no-focus

# parse the new pane id from the JSON response
# the response is: {"result":{"pane":{"pane_id":"w651d456f8444f3-6", ...}}, ...}

# run the agent in the new pane
herdr pane run w651d456f8444f3-6 "pi"

# wait for the agent prompt to appear (adjust match pattern for your agent)
herdr wait output w651d456f8444f3-6 --match "thinking" --timeout 30000

# give it a task
herdr pane run w651d456f8444f3-6 "review the test coverage in src/api/"

do not pipe the agent's output (e.g. pi | tee /tmp/log) when starting — the command text is sent to the pane's shell, not as a pipe target. let the agent's output flow naturally to the pane.

wait patterns

herdr wait output watches for text in a pane's terminal output. useful for servers, builds, logs:

bash
herdr wait output w651d456f8444f3-3 --match "ready on port 3000" --timeout 30000
herdr wait output w651d456f8444f3-3 --match "test result" --timeout 60000
herdr wait output w651d456f8444f3-3 --match "server.*ready" --regex --timeout 30000

herdr wait agent-status watches for the agent status field on a pane. useful for coding agents:

bash
herdr wait agent-status w651d456f8444f3-3 --status done --timeout 60000
herdr wait agent-status w651d456f8444f3-3 --status idle --timeout 60000
herdr wait agent-status w651d456f8444f3-3 --status blocked --timeout 300000

tip: wait output matches against unwrapped recent terminal text. pane width and soft wrapping do not break matches. if you want to inspect the same transcript that the waiter matched, use pane read --source recent-unwrapped.

both commands exit code 1 on timeout.

reading output

bash
# current viewport (what is visible right now)
herdr pane read w651d456f8444f3-3 --source visible --lines 80

# recent scrollback (what was printed recently)
herdr pane read w651d456f8444f3-3 --source recent --lines 50

# without soft wrapping (best for logs)
herdr pane read w651d456f8444f3-3 --source recent-unwrapped --lines 120

# with ANSI colours preserved
herdr pane read w651d456f8444f3-3 --source visible --ansi

workspace management

bash
herdr workspace list
herdr workspace create --cwd /path/to/project
herdr workspace create --cwd /path/to/project --label "api server"
herdr workspace create --no-focus
herdr workspace focus w651d456f8444f3
herdr workspace rename w651d456f8444f3 "api server"
herdr workspace close w651d456f8444f3

tab management

bash
herdr tab list --workspace w651d456f8444f3
herdr tab create --workspace w651d456f8444f3
herdr tab create --workspace w651d456f8444f3 --label "logs"
herdr tab rename w651d456f8444f3:2 "logs"
herdr tab focus w651d456f8444f3:2
herdr tab close w651d456f8444f3:2

close a pane

bash
herdr pane close w651d456f8444f3-3

integrations

install built-in agent integrations for more precise state detection:

bash
herdr integration install pi
herdr integration install claude
herdr integration install codex
herdr integration install opencode
herdr integration status

uninstall:

bash
herdr integration uninstall pi

named sessions

bash
herdr session list
herdr session attach work
herdr session stop work

select a session for CLI commands:

bash
HERDR_SESSION=side-project herdr workspace list

recipes

run a server and wait until it is ready

bash
# split a new pane
RESPONSE=$(herdr pane split w651d456f8444f3-2 --direction right --no-focus)
NEW_PANE=$(echo "$RESPONSE" | python3 -c 'import sys,json; print(json.load(sys.stdin)["result"]["pane"]["pane_id"])')

# start the server
herdr pane run "$NEW_PANE" "npm run dev"

# wait for it to be ready
herdr wait output "$NEW_PANE" --match "ready" --timeout 30000

# read the output
herdr pane read "$NEW_PANE" --source recent --lines 20

run tests in a separate pane and inspect the result

bash
herdr pane split w651d456f8444f3-2 --direction down --no-focus
herdr pane run w651d456f8444f3-3 "cargo test"
herdr wait output w651d456f8444f3-3 --match "test result" --timeout 60000
herdr pane read w651d456f8444f3-3 --source recent --lines 30

check what another agent is working on

bash
herdr pane list
herdr pane read w651d456f8444f3-3 --source recent --lines 80

watch another pane robustly

bash
# inspect what is already there
herdr pane read w651d456f8444f3-3 --source recent --lines 40

# wait for specific output
herdr wait output w651d456f8444f3-3 --match "ready" --timeout 30000

# read unwrapped transcript that the waiter matched
herdr pane read w651d456f8444f3-3 --source recent-unwrapped --lines 40

spawn a new agent and give it a task

bash
# split pane and start the agent
herdr pane split w651d456f8444f3-2 --direction right --no-focus
herdr pane run w651d456f8444f3-3 "pi"

# wait for the agent prompt
herdr wait output w651d456f8444f3-3 --match "thinking" --timeout 30000

# give it a task
herdr pane run w651d456f8444f3-3 "review the test coverage in src/api/"

# wait for it to finish
herdr wait agent-status w651d456f8444f3-3 --status idle --timeout 120000

# read the result
herdr pane read w651d456f8444f3-3 --source recent --lines 100

# close when done
herdr pane close w651d456f8444f3-3

coordinate with another agent

bash
# wait for the agent to finish
herdr wait agent-status w651d456f8444f3-3 --status done --timeout 120000

# read its output
herdr pane read w651d456f8444f3-3 --source recent --lines 100

multi-turn conversation with an agent

bash
# send first message
herdr pane run w651d456f8444f3-3 "review the test coverage"

# wait until idle
herdr wait agent-status w651d456f8444f3-3 --status idle --timeout 120000

# read response
herdr pane read w651d456f8444f3-3 --source recent --lines 80

# send follow-up
herdr pane run w651d456f8444f3-3 "also check src/utils/"

# wait again
herdr wait agent-status w651d456f8444f3-3 --status idle --timeout 120000

# read again
herdr pane read w651d456f8444f3-3 --source recent --lines 80

notes

herdr pane list prints json. parse result.panes[].pane_id for pane ids and result.panes[].focused to find your pane.
herdr pane split prints json. the new pane id is at result.pane.pane_id.
herdr workspace create prints json with result.workspace, result.tab, and result.root_pane.
herdr tab create prints json with result.tab and result.root_pane.
herdr pane read prints text, not json.
herdr pane send-text, herdr pane send-keys, and herdr pane run print nothing on success.
herdr pane read --ansi or herdr pane read --format ansi returns a rendered ANSI snapshot.
herdr pane read --source recent-unwrapped joins soft wraps — use it to inspect the same transcript that wait output matches against.
herdr pane run <id> "cmd" sends text + \r (enter). the command runs in the pane's existing shell process — it does not start a new shell.
herdr pane send-text <id> "text" sends text without enter. use herdr pane send-keys <id> Enter to press enter.
herdr pane send-keys accepts special keys: Enter, Up, Down, Left, Right, Backspace, Tab, etc.
--no-focus on pane split, tab create, and workspace create keeps your current terminal context focused.
use herdr pane read for current output that already exists. use herdr wait output for future output you expect next.
if you are running inside herdr, the HERDR_ENV environment variable is set to 1.
HERDR_SESSION=<name> selects a named session for CLI commands. HERDR_SOCKET_PATH is a low-level override.
there is no built-in delay or --wait flag on pane run or pane send-text. to pace messages to an agent, use herdr wait agent-status --status idle between sends.
pane ids in herdr v0.5.8 are opaque workspace-scoped strings like w651d456f8444f3-2. always re-read ids from herdr pane list — do not hardcode or guess.
do not redirect pane output with > file or | tee. use herdr pane read and herdr wait output instead.
herdr wait output exit code is 1 on timeout.
herdr agent subcommands do not exist in v0.5.8. use herdr pane commands for all pane interaction.

Publisher	ninehills
Repository	`skills`
Skill name	`herdr`
Stars	278
Forks	26
Bundled files	1
Links	Skill source Repository

Herdr

Overview

Installation

Install in TypingMind

Install in another agent

Use it in TypingMind

Works with any AI model

Skill instructions

herdr — agent skill

verified version

concepts

discover yourself

do not redirect pane output to files

pane commands (the only way to interact with panes)

sending messages to a pane

spawn a new agent

wait patterns

reading output

workspace management

tab management

close a pane

integrations

named sessions

recipes

run a server and wait until it is ready

run tests in a separate pane and inspect the result

check what another agent is working on

watch another pane robustly

spawn a new agent and give it a task

coordinate with another agent

multi-turn conversation with an agent

notes

Bundled files

Frequently asked questions

Set up your own AI workspace now

Overview

Installation

Install in TypingMind

Install in another agent

Use it in TypingMind

Works with any AI model

Skill instructions

herdr — agent skill

verified version

concepts

discover yourself

do not redirect pane output to files

pane commands (the only way to interact with panes)

sending messages to a pane

spawn a new agent

wait patterns

reading output

workspace management

tab management

close a pane

integrations

named sessions

recipes

run a server and wait until it is ready

run tests in a separate pane and inspect the result

check what another agent is working on

watch another pane robustly

spawn a new agent and give it a task

coordinate with another agent

multi-turn conversation with an agent

notes

Bundled files

Frequently asked questions

Related AI Skills

Set up your own AI workspace now