Writing Source Nodes

The markdown body of a source node is an instruction set for agents. Write it as a clear, step-by-step runbook.

Structure

A good source node body includes:

1. Context — What this source provides and when to use it
2. Fetch instructions — Which tools to call, with what parameters
3. Interpretation guidance — How to read and present the results
4. Error handling — What to do if calls fail

Example: Jira sprint source

# Current Sprint Tickets
 
This source provides live sprint data from Jira. Before fetching,
check the project-config source for the active project key.
 
## Fetch the sprint
 
Call `jira_get_sprint_issues` with:
- **project**: Use the project key from project-config
- **sprint**: "current"
- **fields**: summary, status, assignee, priority, labels
 
## Interpret the results
 
- Group tickets by status: To Do → In Progress → Done
- Flag any ticket with status "Blocked" or with no assignee
- Highlight tickets with the "urgent" label
- Count total story points and completed story points
 
## If the call fails
 
If Jira is unreachable, note the failure and continue without
sprint data. Do not retry more than once.

Writing tips

• Be specific about parameters — Don’t say “fetch the data.” Say “call tool_name with param: value.”
• Include conditional logic — “If the user is asking about a specific ticket, fetch that ticket. Otherwise, fetch the full sprint.”
• Reference upstream sources — “Read the project key from the project-config source” tells the agent about the dependency chain.
• Keep it readable — The body is markdown. Use headings, lists, and code blocks for clarity.