feat(tools): include structured output in exported .prompt files

[chrisraygill]

Summary

This is the server-side counterpart to the editable output config added to the model runner in genkit-ai/genkit-ui#1919. That PR lets you set structured output in the Dev UI model runner; this PR makes the Export to .prompt action actually include it.

Linked PR: https://github.com/genkit-ai/genkit-ui/pull/1919

Previously createPrompt dropped the generation's output config entirely, so a flow that used structured output lost its schema when exported. The exported front matter now includes the output, with the JSON Schema written as Picoschema.

Before

---
model: googleai/gemini-flash-latest
config:
  temperature: 0.7
---

After

---
model: googleai/gemini-flash-latest
config:
  temperature: 0.7
output:
  format: json
  schema:
    title: string
    servings: integer, number of servings
    ingredients(array):
      name: string
      onSale: boolean
    steps(array): string
---

What changed

• CreatePromptRequestSchema now accepts output (it was being stripped by tRPC input validation before reaching the frontmatter).
• createPrompt maps the output config onto the frontmatter, reading the schema from jsonSchema (generate action shape) or schema (model request shape), and mapping the JSON-producing formats (json, jsonl, array, enum) onto json.
• jsonSchemaToPicoschema converts an object JSON Schema into the compact Picoschema form: required/optional (?), descriptions, enums, scalar and object arrays, nested objects, and additionalProperties wildcards. Non-object top-level schemas (a bare array or scalar) pass through as raw JSON Schema, which Dotprompt also accepts.

Testing

• Unit tests in genkit-tools/common/tests/utils/prompt_test.ts cover the converter and the frontmatter mapping (15 cases).
• Verified end to end against a running Dev UI: exporting a structured-output prompt from the model runner now produces the Picoschema output block above.

Notes

• The frontmatter format field is limited to json/text/media, so jsonl/array/enum export as format: json with the schema preserved. Exact format fidelity could be a follow-up.

Checklist

• PR title follows https://www.conventionalcommits.org/en/v1.0.0/
• Tested (unit tests + end-to-end via the Dev UI export)
• Docs updated (not required)

[gemini-code-assist]

Code Review

This pull request adds support for mapping a generate request's output configuration onto .prompt frontmatter, converting JSON Schema to Picoschema. This includes schema updates, router integration, and helper functions to handle the conversion. Feedback on the changes suggests making the scalarType helper function more robust by safely handling null or undefined schemas to prevent potential runtime crashes.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[MichaelDoyle]

This is really cool.

Is the conversion code useful for other applications, besides dev ui generating dotprompts? I wonder if that code should live somewhere else. e.g. in the dotprompt library, and we just depend on it here.

[MichaelDoyle]

This is interesting -- so there is a mismatch between what can be done in code, and what dotprompt supports?

If so, whenever we do such a conversion, let's leave a comment in the produced dotprompt that explains.
e.g.) # converted from <type>

[pavelgj]

JSON Schema to Picoschema conversion is lossy... Picoschema has a very limited subset of features.

[chrisraygill]

So, how do we think output format set to jsonl should work when converted to picoschema? What if the JSON schema includes unsupported features in picoschema?