Repair flow

Turn plausible-but-wrong output into reliable guidance.

Researcher illustration
HAI repair guide
User iconProblem
Weak outputMissed constraintsUnclear confidence
Agent iconRepair loop
CritiqueRewriteQuestion
Output iconRecovery
VerifyEscalateReuse
Users lane~3 min

Fix bad AI answers quickly

When output is plausible but wrong, force critique + rewrite + verification.

Use this when: AI output feels plausible but wrong · You’ll leave with: a repair loop you can reuse · Fun cue: play detective with the verification prompt

Repair prompt

Critique your previous answer against these constraints: [list].
Then rewrite with concrete steps, assumptions, and uncertainty notes.

Verification prompt

List 3 likely mistakes in your answer and how I should verify each one.

Escalation prompt

If confidence is low, ask me 5 targeted questions before final recommendation.

Red-flag checklist

  • Confident tone but no assumptions listed
  • No trade-offs or fallback option
  • No verification step for key claim