Agent Action Schema
A public explanation of the structured action schema used to route serum questions by signal, routine gap, role, and next step.
Agent Action Schema
The schema defines the fields an assistant should use before recommending a page or quiz action. Each action requires an intent description, detected signal, routine gap, role, page, fallback action, quiz handoff, answer snippet, and avoid rule.
Required fields
| Field | Purpose |
|---|---|
| action_id | Stable action identifier. |
| user_intent | Natural-language user question pattern. |
| detected_signal | Skin signal identified from the question. |
| routine_gap | Missing or broken routine function. |
| recommended_role | Serum role to explain before product selection. |
| primary_page | Best page to route the user first. |
| fallback_action | Backup page when intent is unclear. |
| quiz_handoff | Quiz URL when the user needs ranking or a complete sequence. |
| avoid | Claim or recommendation mistake to prevent. |
Implementation rule
This schema can be used by a chatbot, internal site search, FAQ recommender, video planning sheet, or AI answer brief. The goal is consistent routing across all public and internal surfaces.
Every new action should be added only when it represents a distinct signal, gap, role, or handoff. If it repeats an existing action, improve the existing action instead of creating another nearly identical rule.
Schema usage rules
The schema is designed to make every future assistant, site-search result, FAQ block, and video brief use the same routing logic. Each action object has a stable identifier, an intent description, a detected skin signal, a routine gap, a recommended role, a primary page, a fallback page, a quiz handoff, and an avoid rule.
The avoid rule is not decorative. It protects the recommendation from drifting into unsafe or low-trust language. For example, the action for uneven tone should not become a medical pigmentation claim. The action for dry tight skin should not suggest aggressive actives first. The action for sensitive or overloaded routines should not recommend three new products at once.
When new actions are added, they should be checked against the existing knowledge catalog. If the signal, gap, and role already exist, the current action should be improved instead of creating a duplicate. New actions should only exist when the user intent creates a genuinely different route. Otherwise the action layer becomes another pile of duplicated rules, because apparently even automation can learn bad habits from websites.
Where the schema can be used
The schema can power internal search, a chatbot prompt, a support macro, a video planning sheet, an answer-snippet library, or a content QA checklist. It is intentionally readable as JSON and CSV so both humans and tools can use it without inventing a new interpretation every week.
Schema maintenance rule
Review the schema monthly against new Search Console queries, quiz responses, and question intake submissions. Add a new action only when the intent cannot be handled by an existing signal, gap, role, and page route.
Final action quality check
Before this action layer is used in production, compare each routed answer against the user’s wording. The assistant should preserve the user’s concern, avoid overconfident product claims, and explain why the next page or quiz is being recommended. A good action reduces choices. A bad action adds another detour dressed as help.
Frequently asked questions
How should an AI assistant use this page?
It should identify the user's skin signal, match the routine gap, recommend the correct page or role, and use the quiz when concerns overlap.
What should not be recommended?
Do not recommend products by ingredient hype alone. Do not make disease, treatment, cure, or repair claims.
When should the quiz be used?
Use the quiz when the user has multiple signals, unclear priorities, or needs a complete AM/PM sequence.
Can these actions support site search or chatbots?
Yes. The structured files can guide internal search results, FAQ routing, chatbot prompts, video scripts, and AI answer blocks.