Skip to main content

Prompt History and Snapshots

Prompt history helps you understand what context was generated, when it was generated, and how to reproduce it later. This matters in audit work because you often need to explain which files, diffs, prompts, and repository state informed a review note or LLM response.

What is saved​

Exact fields can vary by build, but prompt history is intended to save metadata such as:

  • Timestamp.
  • Repository reference.
  • Selected file paths.
  • Output format: XML-like or Markdown.
  • Prompt template name or custom instruction summary.
  • Git branch, commit, or diff refs when available.
  • Token estimates.
  • Output hash or context hash when available.

What is not saved by default​

PastePrompt should avoid storing full source contents in history by default. History should prioritize references, metadata, selections, output summaries, and hashes rather than duplicating private repository files.

Check your app settings

If a future build offers full-content history, enable it only when you understand the local storage implications for that repository.

Output hash​

An output hash is a digest of the generated bundle content. It helps answer whether two generated bundles are identical.

Use it to:

  • Compare a recreated bundle with a previous export.
  • Reference a specific context bundle in audit notes.
  • Detect when a repository change, prompt edit, or format change produced different output.

An output hash does not prove that a finding is valid. It only identifies generated content.

Restore previous context​

To recreate a previous context bundle:

  1. Open prompt history.
  2. Find the previous entry by timestamp, repository, or prompt name.
  3. Check the recorded Git branch, commit, or base/head refs.
  4. Restore that repository state locally if needed.
  5. Reapply the saved selection and prompt template.
  6. Generate the bundle again.
  7. Compare the output hash when available.

If files changed or disappeared, the recreated bundle may differ. Record that drift in your review notes.

Why this matters for audit reproducibility​

Audit results are easier to defend when you can trace:

  • Which files were included.
  • Which Git state was reviewed.
  • Which instructions were given to the model.
  • Which bundle was pasted or exported.
  • Whether later context differs from the original context.

This is especially useful for sponsor fix review, follow-up questions, issue validation, and team handoff.