Skip to content
Skills Library
AI & agents · Automation

workflow

/workflow

Deterministic multi-agent orchestration powered by the GitHub Copilot CLI. Use when asked to "run a workflow", "orchestrate", "fan out agents", "run agents in parallel", "spawn subagents", or to decompose a job across many independent agents — audits, code review, sweeps, research, migrations. Instead of working turn-by-turn, write a small JavaScript orchestrator on the fly; each step runs as a headless `copilot -p` child process. Reach for this when a job needs breadth, independent verification, or scale a single context can't hold.

Repository
workflow ↗
Licence
MIT
Last repository change
28.05.2026

Assessment and security

98/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content35 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene13 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
8
Scripts
3
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 28.05.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What to check before installing
  • 3 scripts in this package. Review them before installing, as Claude may execute them.

Scripts: references/route-auth-audit.mjs, tools/smoke.mjs, tools/wf-runtime.mjs

What it does and when to use it

Deterministic multi-agent orchestration powered by the GitHub Copilot CLI. Use when asked to "run a workflow", "orchestrate", "fan out agents", "run agents in parallel", "spawn subagents", or to decompose a job across many independent agents — audits, code review, sweeps, research, migrations. Instead of working turn-by-turn, write a small JavaScript orchestrator on the fly; each step runs as a headless `copilot -p` child process. Reach for this when a job needs breadth, independent verification, or scale a single context can't hold.

Triggers from the description

run a workfloworchestratefan out agentsrun agents in parallelspawn subagents
Original author description

Deterministic multi-agent orchestration powered by the GitHub Copilot CLI. Use when asked to "run a workflow", "orchestrate", "fan out agents", "run agents in parallel", "spawn subagents", or to decompose a job across many independent agents — audits, code review, sweeps, research, migrations. Instead of working turn-by-turn, write a small JavaScript orchestrator on the fly; each step runs as a headless `copilot -p` child process. Reach for this when a job needs breadth, independent verification, or scale a single context can't hold.

How it works

  1. Prerequisites
  2. How to run a workflow
  3. The primitives
  4. Rules (do not skip)
  5. Confidence patterns (compose as needed)
  6. Resume
  7. Reference
Original SKILL.md headings · 82 lines, 528 words
SKILL.md preview
---
name: workflow
description: >
  Deterministic multi-agent orchestration powered by the GitHub Copilot CLI. Use when asked to
  "run a workflow", "orchestrate", "fan out agents", "run agents in parallel", "spawn subagents",
  or to decompose a job across many independent agents — audits, code review, sweeps, research,
  migrations. Instead of working turn-by-turn, write a small JavaScript orchestrator on the fly;
  each step runs as a headless `copilot -p` child process. Reach for this when a job needs
  breadth, independent verification, or scale a single context can't hold.
metadata:
  version: 0.1.0
---

# Workflow: deterministic orchestration with Copilot

You hold the control flow (loops, fan-out, branching) in plain JavaScript; each individual step
is delegated to a fresh headless Copilot agent. The orchestration is deterministic — only the
work inside each `agent()` call is model-powered.

## Prerequisites

- GitHub Copilot CLI installed and logged in (`copilot` once interactively, or `copilot login`).
- Node ≥ 18.

## How to run a workflow

When this skill triggers, do **not** do the task turn-by-turn. Instead:

1. **Write an orchestrator script** (e.g. `workflow.mjs` in the project root), importing the
   runtime that ships with this skill:
   ```js
   import { agent, parallel, pipeline, phase, log } from './.claude/skills/workflow/tools/wf-runtime.mjs'
   ```
   Adjust the path to wherever this skill is installed (e.g. `.github/skills/workflow/tools/…`
   or `~/.copilot/skills/workflow/tools/…`). If unsure, locate `wf-runtime.mjs` first.

2. **Follow the shape: fan out → reduce → synthesize.**
   ```js
   phase('Find')
   const found = await parallel(items.map((x) => () => agent(promptFor(x), { schema: ITEM, label: x.id })))
   const clean = found.filter(Boolean).flatMap((r) => r.items)   // reduce in plain JS — no agent
   phase('Synthesize')
   const report = await agent(`Write a report from:\n${JSON.stringify(clean)}`)
   console.log(report)
   ```

3. **Run it and report the result:**
   ```bash
   node workflow.mjs
   ```
   Progress prints to stderr; the final `console.log` is the answer to relay to the user.

## The primitives

- `agent(prompt, { schema?, model?, label?, retries? })` — one headless Copilot agent.
  With `schema` (a JSON Schema) it returns parsed JSON and retries on invalid output;
  without, it returns raw text. Returns `null` on failure — `.filter(Boolean)`.
- `parallel(thunks)` — run concurrently. **Barrier**: awaits all. Failures become `null`.
- `pipeline(items, ...stages)` — stream each item through stages, **no barrier**. Each stage
  gets `(prev, originalItem, index)`.
- `phase(title)` / `log(msg)` — progress lines on stderr.

## Rules (do not skip)

- **Default to `pipeline()`.** Only use a `parallel()` barrier between stages when a stage
  genuinely needs ALL prior results at once (dedupe/rank across everything, early-exit on total
  count, "compare against the other findings").
- **Each agent is stateless** — a separate process with no memory of the others. Put every bit
  of context it needs into its prompt.
- **Never remove the recursion guard.** The runtime launches children with
  `--no-custom-instructions` so they don't re-load this skill and spawn more workflows.
- **Reduce with plain JS**, not agents — flatten, dedupe, filter, sort are just code.
- **Quota costs are real**: every agent is a full Copilot invocation. Scope `--allow-all-tools`
  down for untrusted tasks, and route cheap stages with `{ model: '<smaller-model>' }`.

## Confidence patterns (compose as needed)

- **Adversarial verify**: for each finding, spawn N skeptics prompted to *refute* it; keep it
  only if a majority survive.
- **Perspective-diverse verify**: give each verifier a different lens (correctness, security,
  performance, reproducibility) rather than N identical ones.
- **Loop-until-dry**: for unknown-size discovery, keep spawning finders until K consecutive
  rounds surface nothing new. Dedupe against everything *seen*, not just confirmed results.

## Resume

Set `WF_JOURNAL=.wf/run.json` before running to cache every `agent()` step. Re-running with the
same journal replays cached results instantly and only re-runs new/changed steps.

## Reference

See `references/route-auth-audit.mjs` for a complete fan-out → adversarial-verify → report
workflow. `tools/smoke.mjs` is a 4-agent end-to-end check that the runtime can drive Copilot.
Files in this package (8)
  • .gitignore 1 KB
  • LICENSE 1 KB
  • README.md 6 KB
  • package.json 1 KB
  • skill.md 4 KB
  • references/route-auth-audit.mjs 3 KB
  • tools/smoke.mjs 1 KB
  • tools/wf-runtime.mjs 7 KB

Versions

RepositoryScoreLicence
This page · alexanderop/workflow98MIT
jeremylongshore/tons-of-skills-marketplace100MIT
SimHacker/moollm100MIT
mlophez/polix-ai100Apache-2.0
stellarie/minimal-agent-kit100MIT
waynesutton/builder-skills100Apache-2.0
claude-world/director-mode-lite100MIT
DawitLemmaBalcha/agent-workflow100MIT
zhinkgit/embeddedskills100MIT

Database snapshot: 13.11.2026 · package version 0f62e22aa5a0 (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100codex-workflow
    Run deterministic multi-step agent workflows in Codex - fan-out, parallel, schema-validated structured output - via the codex-workflow MCP orchestrator (run_workflow), with native in-chat fallbacks. Use when you want Claude-Code-Workflow-style orchestration in Codex: 'run a…
  • 100gh-copilot-cli
    Use when working with GitHub Copilot CLI (`copilot`) — headless and interactive modes, permission flags, AGENTS.md and `.github/copilot-instructions.md`, MCP servers, custom agents, plugins, OAuth device-flow auth, and the bridge to VS Code Copilot Chat. Covers…
  • 84add-other-agents
    Add child agents, connected agents, or other multi-agent patterns to a Copilot Studio agent. Use when the user asks to create a sub-agent, child agent, connected agent, or call another agent.
  • 100dispatching-subagents
    Chooses how to launch subagents from Claude Code (the Agent tool, a Workflow script, or headless claude -p children) and runs a headless fan-out from one background command. Use for "fan this out to 50 agents", "run these tasks in parallel", "launch subagents for each file",…
  • 100neo-sub-agent
    Use this skill when the user asks to design, create, review, or convert sub-agents, custom agents, worker/reviewer/planner agents, background agents, or multi-agent workflows for Antigravity CLI, Codex, Claude Code, GitHub Copilot CLI, or neutral agent blueprints.
  • 100dispatching-subagents
    Chooses how to launch subagents from Claude Code (the Agent tool, a Workflow script, or headless claude -p children) and runs a headless fan-out from one background command. Use for "fan this out to 50 agents", "run these tasks in parallel", "launch subagents for each file",…

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.