agentleFS
davidondrejdavidondrej/skills/skills/herdr/SKILL.md

herdr

Skill4.3k starsChanged 6 days agoMITView on GitHub

Operate and coordinate AI agents in Herdr workspaces in Ghostty. Use when the user mentions Herdr or asks to inspect, message, wait for, or coordinate its agents. Excludes cmux and ordinary terminal sessions.

  • Reads credentials

51 other projects use a version of this, 1 of them edited.

Trending

More talk lately.about the whole repository

Dec 2025: 0 mentionsJan 2026: 0 mentionsFeb 2026: 0 mentionsMar 2026: 0 mentionsApr 2026: 0 mentionsMay 2026: 0 mentionsJun 2026: 0 mentionsJul 2026: 1 mentionAug 2026: 1 mentionSep 2026: 0 mentionsOct 2026: 27 mentionsNov 2026: 5 mentions
Dec 2025Jun 2026Nov 2026
Mentions a month of the davidondrej/skills repository that our weekly crawl found. There is no adoption data for this file yet.

Most recently taken up by

Who's talking about davidondrej/skills

The biggest 5 of 34 mentions linking to the repository that our crawl found. Each line is their own headline.

See all 34

Talked about on GitHub (34 times).

On GitHub · 34

29 more on GitHub

Used with

Projects with it also use better-accessibility, show-me and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • AnasX7Skill

    better-accessibility

    Helps your project comply with accessibility standards and best practices.

    Also in 3 projects that use this

  • AnasX7Skill

    show-me

    Help the user understand the current topic visually with concise diagrams, code-shape sketches, and focused HTML artifacts.

    Also in 5 projects that use this

  • kannaka-labsSkill

    skill-doctor

    Environment diagnostics — check providers, auth, config, hooks, scheduler, and more. Use when: AUTOMATICALLY ACTIVATE when user asks about:. \"doctor\" or \"run doctor\" or \"diagnostics\". \"check my setup\" or \"is everything working\"

    Also in 3 projects that use this

  • GoogleSkill

    wayfinder

    Plan a huge chunk of work — more than one agent session can hold — as a shared map of decision tickets on your issue tracker, and resolve them one at a time until the way to the destination is clear.

    Also in 6 projects that use this

  • AWSSkill

    wizard

    Generate an interactive bash wizard that walks a human through steps only they can perform. Use when provisioning infrastructure, setting up credentials or CI secrets, walking an unfamiliar third-party dashboard, or running a one-off migration or cutover. Don't invoke this for steps the agent can perform itself.

    Also in 5 projects that use this

  • MicrosoftSkill

    writing-for-agents

    Writing documents for agents. Use when creating or editing skills, or modifying AGENTS.md or CLAUDE.md.

    Also in 7 projects that use this

The file

We split this skill into the rules it gives an agent

11 rules. 0 are common, 5 are distinctive, and 6 are very specific to this project.

  • 0

    Common

    Most of the 49 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 6

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run the controlling agent inside Herdr in Ghostty.

    line 12
  2. Distinctive:

    If either fails, tell the user this agent must be launched inside Herdr.

    line 19
  3. Distinctive:

    When coordinating agents, share missing context, prevent duplicate work, and return one combined summary.

    line 48
  4. Distinctive:

    Do not close, rename, move, resize, or reconfigure panes you did not create.

    line 50
  5. Very specific:

    Use named herdr session stop <name> / herdr session delete <name> for destructive operations.

    line 56
  6. Very specific:

    Use an isolated named session, never default, for risky experiments.

    line 60
  7. Very specific:

    Use pane read --format ansi to distinguish placeholder text from human drafts: suggestions are dim (SGR-2) or dark truecolor.

    line 72
  8. Very specific:

    Do not use tput cols for layout in scripts launched through pane run.

    line 87
  9. Distinctive:

    ALWAYS launch agents with auto-approval — an unattended worker can otherwise stall on an unanswered prompt:

    line 92
  10. Very specific:

    Verify launch with herdr agent wait <pane> --status working --timeout MS or herdr wait output <pane> --match <text>, then read.

    line 100
  11. Very specific:

    Use cursor-agent in scripts.

    line 104

“Common” is measured: kept by at least 6 in 10 of the 49 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Herdr

Herdr's CLI controls terminal panes through a local socket; it provides neither native agent-to-agent communication nor shared memory. It works with Pi, Codex, Claude Code, Cursor CLI, and other terminal agents. Native integrations improve detection, status accuracy, and session restoration.

Preconditions

Run the controlling agent inside Herdr in Ghostty. Verify access and workspace scope:

test "${HERDR_ENV:-}" = "1"
test -n "${HERDR_WORKSPACE_ID:-}"

If either fails, tell the user this agent must be launched inside Herdr; do not claim access to other agents.

One-time installation:

npx skills add ogulcancelik/herdr --skill herdr -g

Inspect, message, wait, read

Append --session <name> to every Herdr command. HERDR_SESSION alone can silently select an already-running server. Discover pane IDs; never guess them.

herdr pane list --workspace "$HERDR_WORKSPACE_ID" --session <name>
herdr pane read <pane-id> --source recent-unwrapped --lines 200 --session <name>

# For TUI agent composers, separate text from Enter
herdr pane send-text <pane-id> "Check the failing tests and report back." --session <name>
sleep 1
herdr pane send-keys <pane-id> enter --session <name>

# Confirm submission, then wait for completion (idle if the integration uses it)
herdr agent wait <pane-id> --status working --timeout 120000 --session <name>
herdr agent wait <pane-id> --status done --timeout 120000 --session <name>
herdr pane read <pane-id> --source recent-unwrapped --lines 200 --session <name>

Stay in the current workspace. Read before messaging; send nothing for inspection-only requests. Confirm submission through a native status transition to working/blocked after Enter, never merely changed pane content. Wait for done, idle, or blocked, then read again; do not assume the prompt was acted on. Corroborate idle with pane text as described below.

When coordinating agents, share missing context, prevent duplicate work, and return one combined summary.

Do not close, rename, move, resize, or reconfigure panes you did not create. Create or close panes only when the user explicitly asks. “Use the Herdr skill” means execute the CLI workflow, not merely explain it.

Operational details

Sessions and targeting

  • Use named herdr session stop <name> / herdr session delete <name> for destructive operations. Never use herdr server stop; it targets the ambient server.
  • Targets are <session>:<pane-id>. Pane IDs contain a colon (w1:p2); split on the first colon only.
  • CLI calls do not auto-start a server. Start headless with herdr server --session <name>.
  • Managed processes receive HERDR_ENV=1 and HERDR_PANE_ID. Inside nested tmux, $TMUX wins: treat the pane as tmux.
  • Use an isolated named session, never default, for risky experiments. Re-check herdr session list --json --session <name> immediately before stopping or deleting.

Input

  • pane send-text types without submitting. pane run sends text + Enter and works at shell prompts, but TUI composers (Claude Code, Cursor CLI) may swallow Enter as part of a paste. Use separate send-text, ~1s pause, then send-keys <pane> enter calls for TUIs.
  • C0 control bytes (e.g. ASCII 0x1f) trigger terminal actions and can erase typed text. Use U+2063 INVISIBLE SEPARATOR for invisible text markers.
  • Slash-command autocomplete may consume the first Enter to close the popup or fill an argument placeholder. escape dismisses the popup while preserving text.

Reading and status

  • pane read --lines N returns empty output if N is below viewport height. Request at least 200 lines and at least the viewport height; trim locally with tail -n N.
  • pane get .cwd is fixed at creation; .foreground_cwd is live.
  • Use pane read --format ansi to distinguish placeholder text from human drafts: suggestions are dim (SGR-2) or dark truecolor; typed input renders normally. Plain text cannot reliably distinguish them.
  • herdr agent get <pane> reports native working/idle/done/blocked/unknown status; prefer it to regex guesses. Long-running foreground tools may still report idle; check busy banners such as "esc to interrupt" before treating the pane as free or stale.
  • Native waits: herdr agent wait <pane> --status <s> --timeout MS, herdr wait agent-status <pane> --status <s> --timeout MS, and herdr wait output <pane> --match <text>.
  • Socket protocol >=16 offers pane.agent_status_changed and pane.output_matched events. Prefer events, with polling as a backstop.
  • Scripts can register via pane report-agent and report idle/working/blocked.

Workspace and tab lifecycle

  • Labels are not unique. Check duplicates yourself; find-by-label adopts the first match. Unlabeled workspaces display their cwd basename, which can collide with an explicit label.
  • workspace create seeds tab 1. Closing the last tab deletes the workspace; closing a tab's only pane closes the tab.
  • Workspace/tab creation respects --no-focus, except the first workspace in an empty session. pane split --no-focus still shrinks the host viewport: focus and geometry are separate.
  • Named-session restarts preserve IDs and labels, but not processes or agent registrations. Restored panes are fresh shells (agent_not_found), not live duplicates; close and replace them only within the pane permissions above.

Other details

  • Do not use tput cols for layout in scripts launched through pane run; it can report a stale default of 80.
  • herdr integration install <harness> (claude, codex, cursor, pi, ...) enables native status detection. herdr notification show <title> displays an alert.

Launch agents only when the user asks

ALWAYS launch agents with auto-approval — an unattended worker can otherwise stall on an unanswered prompt:

  • Cursor CLI: cursor-agent --yolo "task" (alias for --force)
  • Codex CLI: codex --yolo "task"
  • Claude Code: claude --dangerously-skip-permissions "task"

First-run trust dialogs may still appear; inspect the pane after launch. Install each integration once for native status detection.

Verify launch with herdr agent wait <pane> --status working --timeout MS or herdr wait output <pane> --match <text>, then read. Never use sleep N && pane read as launch verification; the short input-submission pause above serves a different purpose.

Cursor CLI

Use cursor-agent in scripts. agent is an alias/new docs name; the user's interactive cur shorthand may expand to cursor-agent --yolo, but aliases do not expand in scripts.

Discover available model slugs with cursor-agent --list-models before launching into an existing shell pane:

herdr pane run <pane-id> "cd <worktree> && cursor-agent --model <verified-model-slug> --yolo 'fix the failing tests'" --session <name>
  • Interactive: cursor-agent "task", or no argument for an empty session. Headless: cursor-agent -p "task" --output-format text|json|stream-json.
  • Permissions: --force / --yolo skips per-command approval; --sandbox enabled|disabled controls sandboxing. Script auth: CURSOR_API_KEY.
  • Model: --model <slug> at launch, /model in-session. Slugs vary by version/account; never invent them from memory.
  • Effort: no --effort flag. Use full slugs ending in -low, -medium, -high, or -xhigh (e.g. gpt-5.3-codex-xhigh, claude-opus-4-8-thinking-high). -fast controls speed, not effort. Despite --help, model[effort=high] is unsupported.
  • Some builds silently drop --model reasoning suffixes. Verify with /model after launch; --disable-auto-update preserves a working build.
  • Resume: cursor-agent resume (latest), --resume <chatId>, or cursor-agent ls to list.

Direct shortcuts

Most defaults use Ctrl+B. Direct Ctrl+Alt shortcuts rarely conflict. Add to ~/.config/herdr/config.toml:

[keys]
new_workspace = "ctrl+alt+n"
workspace_picker = "ctrl+alt+w"
goto = "ctrl+alt+g"
new_tab = "ctrl+alt+c"

Apply with:

herdr server reload-config --session <name>

To keep prefixed shortcuts with a different prefix, set prefix = "ctrl+a" under [keys].

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.