Template parameters
Every prompt template can carry parameters — flat name → value entries that act as the default runtime knobs for that template. The runtime modal (e.g. the Generate content wizard, the inline modal) and the Test on product pane both read these values when they open. Suggestions accepted from the AI assistant chat or the learning flow can also write straight into them.
Parameters are managed from a dedicated tab on the prompt detail page.
Where to manage them
Section titled “Where to manage them”- Open any template from AI Assistant → Prompt templates.
- Switch to the Template parameters tab.
- Click Edit at the top of the page to make the form interactive.
- Adjust the values for this template and click Save.
The help text on the card reads exactly:
Default runtime parameters for this template. Forks inherit each parameter from the parent template; an edit creates an own row for this template only. The runtime modal and the test pane read these values on open.
Available parameters per task type
Section titled “Available parameters per task type”The set of parameters surfaced depends on the template’s task type — there is no point showing a “length max” field on a translate template. The plugin resolves the baseline list from PromptDefaultsLibrary (with a JS mirror in helper/prompt-defaults-library.js):
| Task type family | Parameters |
|---|---|
generate_content.* | sourceMode (model / web_search / url_fetch), webSearchContextSize (low / medium / high), customInstructions, plus per-field fieldConfig.<field>.lengthMin, fieldConfig.<field>.lengthMax, fieldConfig.<field>.formatSpec, fieldConfig.<field>.customInstructions for name, description, metaTitle, metaDescription, keywords. |
improve.* and inline.improve | preset (default seoOptimized), customInstructions, fieldsList (null = all fields). |
translate.* and inline.translate | fieldsList, customInstructions. |
unify.* and inline.unify | formatSpec, customInstructions. |
inline.describe_from_url | preset, unifyFormatting, customInstructions. |
generate_seo.*, generate_alt_text.*, extract_measures, extract_properties | customInstructions only. |
Other task types (image edit, image finder, media index, category and keyword suggestions, meta-prompts) don’t expose template parameters today. The category and keyword assistants take their runtime knobs — number of categories, keyword min/max, source fields — from the modal each time you open it, rather than from a stored parameter row.
How values are resolved
Section titled “How values are resolved”When a runtime modal or test pane opens, the plugin computes the effective value for each parameter using a 4-layer fallback:
- Own row — a parameter row attached directly to this template wins.
- Ancestor chain — if there’s no own row, the plugin walks the fork chain (parent → grandparent → …, up to 16 levels). The first ancestor that has a row for that name wins.
- Task-type baseline — if neither this template nor any ancestor has a row, the
PromptDefaultsLibrary::forTaskType(...)baseline is used. - Test pane override — finally the user can override any value ad-hoc inside the Test on product pane; that override is per-call and never persists.
Each parameter resolves independently by name — there’s no deep-merge. A fork can override just fieldConfig.description.lengthMax and inherit every other parameter from its parent.
The Template parameters form surfaces the source for each row — own values are highlighted and offer a Reset to inherited action. inherited from parent and baseline values are read-only until you edit them (writing creates an own row).
Forks: inheriting and overriding
Section titled “Forks: inheriting and overriding”Parameters are the most lightweight way to customise a fork. If a fork only needs longer descriptions but inherits the same tone, the system + user prompts can stay identical and you just override fieldConfig.description.lengthMax. The fork chain walks until it finds an own row, so:
- Edit the default template’s parameters → all forks of that type inherit the new value (unless they already have their own row).
- Edit a fork’s parameter → only that fork and its descendant forks see the change.
- Reset to inherited on any row → that row is deleted and the value falls back to the parent / baseline again.
AI suggestion → parameter
Section titled “AI suggestion → parameter”When you refine a prompt with the AI assistant chat, the model can propose parameter changes — for example a longer max length — in a Suggested adjustments panel alongside its prompt edits. Clicking Apply on a row writes that single parameter, the same one-row update the Template parameters form makes.