Skip to content

Resilience and Repairs (Overview)

Genie Repair helps you recover when generated output is incomplete, inconsistent, or likely to fail in your real script/workflow.

  • Analyzes generated content against the task and available context.
  • Detects likely issues before you apply changes.
  • Produces a repaired version when a safe fix is possible.
  • Uses model-assisted repairs for complex cases that need deeper interpretation.
  1. Analyze: Genie inspects the result, your prompt intent, and available context.
  2. Classify: Genie flags warnings, assumptions, or missing pieces that can affect reliability.
  3. Repair: Genie applies direct fixes when possible, or uses model-assisted repair for harder issues.
  4. Return: You get updated output plus clear indicators of what changed and what still needs review.
  • The issue requires semantic understanding, not just formatting cleanup.
  • Multiple constraints must be reconciled (for example, intent vs. context mismatch).
  • A safe deterministic fix is not enough to produce a reliable result.

What you should do before applying repaired output

Section titled “What you should do before applying repaired output”
  • Review warnings and assumptions shown in chat.
  • Confirm referenced sources when web-backed context is used.
  • Validate the repaired result in your real script flow before broad rollout.
  • Business logic is ambiguous or incomplete.
  • Required metadata/context is missing or outdated.
  • You need a specific implementation style that was not stated in the prompt.

Example of model-assisted repair result with change summary.

Section titled “Example of model-assisted repair result with change summary.”

Example of a manual auto-repair + notification AutomaticRepairsExample.png