Skip to main content
Writing a good support prompt from scratch is slow. A pack is a starting point for one kind of agent: a starter prompt and a set of starter skills. In this part you apply the Support pack to the order-support agent you connected in Part 2, fill in the starter prompt’s blanks, and then make one of its skills your own. Your code doesn’t change.

​
1. Look at the Support pack

Open the Support pack. It has:
  • A starter prompt, about fifteen lines: what the agent may and may not do, a line for its tools, and its tone. It has blanks written in angle brackets, like <product>, for the parts only you know.
  • Eight starter skills for the steps of a support conversation: triage, billing disputes, missed service levels, reply format, follow-ups, account deletion, escalation, and incident updates. order-support already has three of them from Part 2.

​
2. Apply it to order-support

The Free plan includes one agent, so apply the pack to the agent you have rather than creating a second one. Use the CLI or the dashboard; both make the same change.
Needs decimalai 0.26.0 or later (pip install -U "decimalai[openai-agents]"). Run it from the project, where .env has your DECIMAL_API_KEY:
It shows what will change and asks first (example output):
--dry-run shows the change and stops, and --yes applies without asking. The CLI reference has the other options.
What applying did:
  • The starter became the prompt’s next version, v2, marked as coming from the pack and labelled Applied the Support pack. Your v1 stays in the Prompt tab’s history; you can go back to it at any time.
  • Five skills were added: sla-breach-response, support-reply-format, csat-followup-request, mohitagw15856-escalation-tree and incident-status-comms. The three the agent already had stayed: billing-dispute-resolution, account-deletion-verification-policy and diagnostic-triage-questions.
  • It is one change on the agent’s Versions page: a new setup version holding both the prompt and the skills.
Applying the same pack again changes nothing; the CLI says Nothing to apply. If you had pinned the agent to an older prompt version, the pin stays: the starter is saved as a new version but not served, and the CLI says so.

​
3. Fill in the blanks

On the Prompt tab, edit the new version:
  • Replace <product> with Acme's online store.
  • Replace the Tools line’s blank with lookup_order, the one tool your code attaches:
Save. That makes v3.

​
4. Run the same code

Run main.py from Part 2, unchanged:
The agent now runs on the saved prompt, v3, with eight skills on its menu. Your code still passes fallback=, but the agent has a saved prompt now, so the fallback runs only when DecimalAI can’t be reached at startup.

​
5. Make one skill your own

The starter skills are other people’s text. To change one, make an editable copy for your organization. Here you’ll add a refund rule to billing-dispute-resolution.
  1. Open billing-dispute-resolution and choose Fork editable. Pick order-support as the agent, then click Create editable copy. The copy keeps the skill’s name and replaces the registry original on order-support; the original is unchanged.
  2. On your copy’s Settings tab, add a line to Body (Markdown):
  3. Click Save Changes, then Publish. Editing the body makes a new version of your copy.
Run main.py again. The double-charge question is about a $129 charge, so when the model loads the skill, the reply should now say the charge has been escalated rather than promise a refund. No redeploy: the agent’s skills live on DecimalAI, and its next runs pick up your copy.

​
Optional: start a project from the agent

decimalai init writes a runnable file for an agent that already exists in your workspace, with its skills and prompt loaded at run time:

​
Optional: a frozen copy instead

decimalai starter support downloads a local Support project with the pack’s prompt and skills copied into it. That copy is frozen: it needs no DecimalAI SDK or account, and it sends no traces, so nothing you change in the dashboard reaches it. Support agent from skills walks through that path.

Next: 4. See what it did

Read order-support’s traces: the skills it was offered and loaded, the tools it called, and the prompt and skill versions each run used.