Scrydon
SDKsPacks & Authoring SDK

Author a Pack with an AI Coding Agent

No coding experience required — describe the Process Flow, Ontology, or Data Source you want in plain language, and let an AI coding agent like Claude Code, Cursor, or GitHub Copilot do the technical work of building the pack for you

Packs are built using code, not a visual editor. See Packs & Authoring SDK if you want the technical details. This page doesn't change any of that — it shows you how to get a pack built anyway, without writing or reading any code yourself, by describing what you want to an AI coding agent — a tool like Claude Code, Cursor, or GitHub Copilot's Agent mode — that writes and runs that code for you.

If you can describe, in your own words, the process you want — what steps it goes through, who's responsible for each one, and what needs approval — an AI coding agent can turn that description into a working pack. You won't write any code. The only thing it can't do for you is the very last step: uploading the finished pack, which happens in your web browser, in Settings → Platform → Packs.

What is an AI coding agent?

An AI coding agent is an AI assistant that goes beyond a chat window — it can also read and write files, and run small programs on your computer. So instead of just replying with an answer, it can actually go and do the work: writing the pack's files, running the tools that check and build them, and fixing its own mistakes if something goes wrong. It's the same process a developer would follow by hand — you just don't have to be the one doing it.

Several tools work this way, including:

  • Claude Code — Anthropic's agent, available in a terminal, IDE, desktop app, or browser.
  • Cursor — a code editor with a built-in agent.
  • GitHub Copilot, in VS Code's Agent mode.
  • OpenAI Codex — runs in a terminal or IDE.
  • Gemini CLI — Google's terminal-based agent.

Use whichever one you already have, or already trust for this kind of work — every step below works the same way regardless of which agent you pick.

Someone needs to actually have one of these agents installed to follow this guide — that's either you, or a colleague you hand this page to. If your organization doesn't let everyone install this kind of software, ask whoever on your team already has one (for example, someone in IT or engineering) to run through the steps below with you, using the prompt template further down.

Before you start

  1. An AI coding agent, installed and signed in. Any of the tools listed above works — follow that tool's own setup instructions; it typically takes a few minutes.
  2. Access to Scrydon's Settings → Platform → Packs, so you can do the upload at the end. You'll need admin rights on the Scrydon organization you're uploading to — if you're not sure whether you have this, ask your Scrydon administrator.

That's genuinely everything. You don't need to know how to code, and you don't need to understand any of the technical terms that show up on this page — your agent handles all of that.

Your agent may also need one small supporting program on your computer to actually build the pack. If it's missing, a capable agent will notice and offer to install it the first time you ask it to build something — just say yes when it asks for permission.

Step by step

All you need to do here is create one new, empty folder to hold this pack, and get your agent running inside that folder. Exactly how you do the second part depends on which kind of agent you're using.

If your agent runs in a terminal (Claude Code, OpenAI Codex, Gemini CLI, and similar) — a terminal is a plain text window for typing instructions to your computer. Pick your operating system below for exact click-by-click instructions.

  1. Open File Explorer and go to wherever you'd like to keep this pack (e.g. Documents).

  2. Right-click an empty area → NewFolder. Name it customer-onboarding-pack, then open it.

  3. With that folder open, right-click an empty area inside it and choose Open in Terminal (Windows 11), or hold Shift and right-click → Open PowerShell window here (Windows 10).

  4. A window will open with a blinking cursor. Type the line for your agent and press Enter (don't type all three — just the one that matches your agent):

    claude   # Claude Code
    codex    # OpenAI Codex
    gemini   # Gemini CLI

Prefer typing over clicking through File Explorer? Open PowerShell or Windows Terminal from the Start menu and type these lines, pressing Enter after each one — using your agent's own command in place of claude:

mkdir customer-onboarding-pack
cd customer-onboarding-pack
claude

Using an IDE-integrated agent instead — like Cursor or GitHub Copilot in VS Code? You don't need a terminal for this step. Open the folder you created as your project in the IDE (usually File → Open Folder…), then open the agent's chat or agent panel — for example, in VS Code with Copilot, open the Copilot Chat panel and switch the mode dropdown to Agent. Once it's ready to receive instructions, skip ahead to the next step.

If your agent's start command isn't recognized, it either isn't installed yet, or the terminal window needs to be closed and reopened after installing it. Go back to Before you start above.

Now that your agent is running, describe the process you want — as if you were explaining it to a new colleague. There's a fill-in-the-blanks message you can copy for this in the prompt template section further down this page. The more concrete you are about each step, who's responsible for it, and what needs sign-off, the less back-and-forth you'll need.

From here, it's your agent's job, not yours. It will set up the tools it needs, write out all the pack's files behind the scenes, and then check its own work — twice: once for obvious mistakes, and once by actually building the finished pack. If either check turns up a problem, a capable agent reads the error itself and fixes it, without you needing to understand what went wrong. This can take a few minutes and a few tries; that's completely normal, and your agent should keep you posted on what it's doing.

Before you trust the result, ask directly: "Explain, in plain language and with no code, what steps, tasks, and approvals this pack creates." Read the answer and confirm it actually matches what you asked for. This is the one review step that's genuinely yours to do — no technical knowledge required, just checking that it understood you correctly.

When your agent is done, it will tell you the name of the file it created — it will end in .tar.gz. This is the only step that happens in your browser, and the only step your agent genuinely can't do for you — it has no way to log into your Scrydon account. See Uploading your pack below for the click-by-click walkthrough.

Once it's uploaded, a workspace admin adds the template to an environment from the Marketplace — a few clicks in Scrydon's UI. After that, anyone can start a new run of it from Process Flows, just like any other template.

Prompt template

You don't need to understand the text below — it's written for your AI coding agent, not for you, and it looks more technical than anything you'll actually have to do. Just copy the whole thing, paste it into your agent from Step 1, and replace the parts in [brackets] with your own description, in your own words.

I want to build a Scrydon Pack containing a Process Flow (and an Ontology if
one is needed) using @scrydon/sdk-authoring. I don't write code myself —
please do the implementation, run the build, and fix any validation errors
yourself. Explain what you built in plain English when you're done, before
I upload it.

Ground yourself first — fetch these docs pages before writing anything:
- https://docs.scrydon.com/sdks/authoring for the overall pack lifecycle
  (define → validate → bundle → upload).
- https://docs.scrydon.com/sdks/authoring/process-flows for the Process
  Flow SDK (stages, tasks, personas, action templates).
- If the process needs a data model, also fetch
  https://docs.scrydon.com/sdks/authoring/ontologies.
- https://docs.scrydon.com/sdks/authoring/process-flows/examples for a
  complete, working pack shape to follow.

Here is the process I want, described in plain language:

[Describe the process end to end. For example:
 "A customer-onboarding kickoff. It has three stages: Discovery, Setup,
 and Go-Live. In Discovery, the Customer Success Manager fills out a
 checklist and uploads the signed contract. Discovery moves to Setup once
 that's done. In Setup, the Customer Champion approves a rollout plan
 document — this needs a manual approval, not automatic. Once approved,
 Go-Live starts and just has one task: a final confirmation checklist
 assigned to the CSM."]

Roles involved: [e.g. "Customer Success Manager", "Customer Champion"]

When you're done:
1. Run `bunx @scrydon/sdk-authoring pack validate . --manifest-only`, then
   `bunx @scrydon/sdk-authoring pack build .`, and fix anything that fails.
2. Tell me the exact filename of the built .tar.gz.
3. Summarize the stages, tasks, and approvals you created — no code, plain
   English — so I can confirm it matches what I asked for before I upload it.

Already have a pack you like the shape of (yours, or one of the worked examples)? Tell your agent to use it as a starting point and describe only what should be different — that's usually faster and more predictable than starting from a blank description.

Uploading your pack

This is the one step you do yourself, in your browser — here's exactly what it looks like.

In Scrydon, open Settings, then Platform, then Packs. You'll land on a page listing every pack already installed in your organization — it may well be empty the first time you do this.

Near the top of the page there's a button labeled Upload pack (or Upload your first pack if the list is empty). Click it, and a panel slides in from the right side of the screen.

Drag the .tar.gz file your agent created straight into the box that says "Drop a pack file here, or click to pick" — or click the box and browse to the file instead.

Scrydon shows a short summary of what's inside the file — broadly the same thing your agent already told you it built in Step 4. Nothing is installed yet at this point; you're just previewing it.

If your pack also includes a Workflow (not just a Process Flow), you'll additionally be asked to pick which workspace environment to install it into — choose the one where you want it to actually run.

You'll see a line that reads "I understand this pack is unsigned and will be audit-logged as such." — tick it. This just means the pack came from your own AI-agent session rather than a signed, published release, so Scrydon keeps a record of who installed it and when; that's expected and nothing to worry about for a pack built for your own organization. Click Upload to finish.

Your pack now appears in the list on the Packs page. From there, a workspace admin adds it to an environment from the Marketplace — after that, it's ready for anyone to use from Process Flows.

Updating a pack later

Wanting to change something later — add a step, tweak who's responsible for a task, fix a typo — doesn't mean starting over. Go back to your agent, in the same folder you used before, and describe the change the same way you described the process the first time.

A few things worth knowing before you do:

  • Ask your agent to bump the pack's version number (for example, from 1.0.0 to 1.0.1) before it rebuilds. Scrydon tells one version of a pack apart from another by that number, not by the filename — if it doesn't change, Scrydon won't recognize the new file as an update.
  • Upload the new file the same way — Settings → Platform → Packs → Upload pack, exactly like the first time.
  • What happens next depends on what's in your pack:
    • If it's just a Process Flow, the new version goes live automatically the moment it's uploaded — there's nothing else to click.
    • If it also includes an Ontology (a data model) and you changed something that already existed, the pack will show as Update available in the list — click into it, then click Update pack to apply that part.

For the full picture on how updates behave for every kind of pack content — including integrations, data sources, and workflows — see Pack Version Management.

What your agent does, and what's still yours to do

  • Your agent handles: writing every file, running the technical build and validation steps, and fixing its own mistakes. You never write code, never learn a file format, and never type a build command yourself.
  • Still yours to do: uploading the finished file in your browser (your agent has no way to log into Scrydon), turning the template on for your workspace via the Marketplace, and — most importantly — checking that what it built actually matches what you asked for. Always read the plain-language summary from Step 4 before you upload.
  • Your agent cannot touch your Scrydon account at all. It only ever works with files on your own computer and the small tools it installs there.

Where to next

These go deeper into how packs actually work — nothing here is required to get your first pack built, but they're useful if you're curious or want to fine-tune something later.

On this page

On this page