Orchestrate Herdr
Fan a PRD or parent issue into one Herdr worker tab per open sub-issue, each running your chosen coding CLI, while you monitor test-backed completion from the orchestrator tab.
Overview
Orchestrate-herdr is an agent skill for the Build phase that fans a PRD into Herdr worker tabs running a chosen coding CLI and monitors them for test-backed completion.
Install
npx skills add https://github.com/devarfeen/agent-skills-kit --skill orchestrate-herdrWhat is this skill?
- Parameterized intake: PRD_URL and CODING_CLI resolved from args or explicit user prompts—no guessing
- Orchestrator-only role: do not implement features or close the orchestrator tab
- Non-negotiables: Herdr-managed tabs only—no panes, nested agents, or internal sub-agents
- Fans open sub-issues from a PRD into per-issue worker tabs running a chosen coding CLI
- Monitors workers for test-backed completion; prompt is fixed—run verbatim after header substitution
Adoption & trust: 1 installs on skills.sh; trending (+100% hot-view momentum).
What problem does it solve?
You have a PRD with many open sub-issues but only one chat session, so parallel implementation and completion tracking do not scale.
Who is it for?
Builders using Herdr who want a repeatable, verbatim orchestration prompt to parallelize PRD sub-issues across coding CLI workers.
Skip if: Users outside Herdr (no HERDR_ENV=1), single-ticket fixes, or workflows that require nested sub-agents or custom rewritten orchestration logic.
When should I use this skill?
Running inside herdr (HERDR_ENV=1) when the user wants to fan a PRD out to per-issue workers with a chosen coding CLI.
What do I get? / Deliverables
Each open sub-issue gets a dedicated Herdr worker tab under a single orchestrator that watches for test-backed done states without nested agents.
- One Herdr worker tab per open sub-issue running the specified CLI
- Orchestrator monitoring state toward test-backed completion
Recommended Skills
Journey fit
Multi-issue implementation orchestration sits in Build because it coordinates parallel coding agents against a product backlog, not distribution or production ops. Herdr tab orchestration is agent-runtime tooling—how you launch and supervise CLI workers—so agent-tooling is the canonical shelf.
How it compares
Herdr-specific parallel tab orchestration, not a generic GitHub Issues bot or a single-agent implementation plan skill.
Common Questions / FAQ
Who is orchestrate-herdr for?
Developers running inside Herdr who manage PRDs with multiple open sub-issues and want one orchestrator tab plus per-issue coding CLI workers.
When should I use orchestrate-herdr?
During Build when a PRD URL is ready, sub-issues are open, and you need to launch codex, claude, or another CLI in separate Herdr tabs under HERDR_ENV=1.
Is orchestrate-herdr safe to install?
It drives browser tabs and coding CLIs in your Herdr workspace; review the Security Audits panel on this page and your CLI permissions before running at scale.
SKILL.md
READMESKILL.md - Orchestrate Herdr
# Orchestrate herdr This is a tested, working orchestrator prompt. The **only** things that change between runs are the **PRD URL** and the **coding CLI** to launch. Do not edit, summarize, or "improve" the prompt — run it exactly as written below. ## Intake Resolve two values before running the prompt: - `PRD_URL` — the PRD or parent issue URL whose open sub-issues become workers. - `CODING_CLI` — the coding CLI command to run in each worker tab (e.g. `codex`, `claude`). Source order: 1. Use values passed as skill args (a URL plus a CLI name, or `PRD_URL=... CODING_CLI=...`). 2. For anything still missing, ask the user. Do not guess. Do not proceed until both are set. Then set the two header lines below to the resolved values and execute the prompt **verbatim**. --- PRD_URL: [PRD_URL] CODING_CLI: [CODING_CLI] You are the **orchestrator**. Do not implement. Do not close this tab. ## Non-Negotiables * Use the existing `herdr` workspace/session only. * Use Herdr-managed tabs only. * No panes. * No internal sub-agents. * No nested agents. * No nested coding sessions. * Stay in the same folder as this orchestrator tab. * Never `cd` into task folders, issue folders, worktrees, or any other folder when creating worker tabs. * Create workers only as new Herdr tabs inside the current workspace. * Do not launch `CODING_CLI` from inside another `CODING_CLI`. * In each worker tab, run only `CODING_CLI` from the original current folder. * Do not pass prompts as launch arguments/flags. * Do not leave prompts staged or unsent. ## Orchestrator Workflow 1. Load required skills. 2. Read `PRD_URL`. 3. Find open sub-issues / linked child issues. 4. Save current Herdr workspace/session ID. 5. Save current working folder. 6. For each open issue, create one Herdr-managed worker tab in the saved workspace and same folder. 7. Name each tab: `[CODING_CLI] - GH #[ISSUE_NUMBER]`. 8. Save each tab ID immediately after creation. 9. Use only saved tab IDs for prompt submit, read, monitor, and follow-up. 10. Never rely on active tab, latest tab, visual order, or guessed tab. ## Worker Launch For each saved worker tab: 1. Confirm tab ID, workspace/session ID, and folder match the saved orchestrator context. 2. Run only `CODING_CLI`. 3. Wait 30 seconds for CLI readiness. 4. Paste the issue prompt into that same saved tab. 5. Submit the prompt. 6. Confirm the first response is visible. A worker is not launched until its tab ID is saved, prompt is submitted, and first response is visible. ## Worker Prompt Do not send the full PRD to workers. Do not send identical prompts to workers. Each worker receives only its assigned issue. ```md GH_ISSUE: #[ISSUE_NUMBER] GH_ISSUE_URL: [ISSUE_URL] Work only on this GitHub issue. Infer project/repo context from the assigned issue. Use `/tdd` / `$tdd` and `/caveman` / `$caveman`. Do not work on the full PRD. Do not redo PRD orchestration. Do only issue-level discovery required for this issue. Avoid unrelated changes. Report back to the main orchestrator/Herdr when completed, errored, or blocked. Completion requires test evidence. ``` ## Monitoring Monitor all saved Herdr tab IDs every 1 minute. Do not accept completion without test evidence. Do not add `BLOCKER:`, `AFK:`, or `HITL:` to issue titles. Use issue order, dependency notes, comments, `ready-for-agent`, and `ready-for-human` labels. Report only: workspace/session ID, folder, tab ID map, assigned issues, workers/status, test evidence, blocked/error issues, completion report per issue.