← knowledge.oriz.in

Always parse 'Other' answers in MCQs for additional context beyond the literal question

rule rulecommunicationmcqaskuserquestionparsingcontextstt-friendly

Parse 'Other' answers for additional context

Rule

When the user answers an multi-choice question prompt MCQ with free-text "Other" (especially via STT), that text often carries:

  1. The literal answer to the question being asked
  2. New directives, constraints, preferences for OTHER topics
  3. Corrections to past decisions
  4. Requests for follow-up grilling
  5. Pasted context (schemas, lists, URLs)

The agent MUST:

  1. Parse the entire free-text, not just match a phrase to the option labels.
  2. Acknowledge each extracted directive explicitly in the response.
  3. Write durable directives to knowledge/ in the same turn (per keep-knowledge-fresh).
  4. Apply non-durable directives to the immediate work.
  5. Still record an answer to the original question (use best inference from the text).

Never silently drop side-context that doesn't fit the original question.

Example

Q: "Which subdomains to wire first?" User's "Other" text: "auth.oriz.in, packages.oriz.in. And every AI provider should be OpenAI-compatible — that's a new rule. Grill me on it."

Agent must:

  1. Record auth.oriz.in + packages.oriz.in as the answer.
  2. Acknowledge "OpenAI-compat for all providers" as a new directive.
  3. Write a new rule knowledge/rules/interaction/openai-compat-for-all-ai-providers.md.
  4. Schedule follow-up MCQs to grill on the new directive.

Why

STT users tend to package multiple thoughts in one utterance. The agent's job is to extract every actionable signal, not just the one that fits the current question template.

This pairs with [[rules/communication-stt-friendly]] — STT input is noisy AND multi-topic.

Cross-refs