AddThisFeature

Prompt Versioning

Tie every AI output to the exact prompt version that produced it.

involved Developer Experience

What it adds

Immutable, numbered versions of each prompt, with the run configuration recorded and every output stamped with the version used.

What your agent is told to do

5
  1. 1

    Make every publish create a new immutable version rather than overwriting the previous text. Editing history in place destroys the only record of what produced last month's outputs.

  2. 2

    Capture the whole run configuration with each version, not just the wording: which model tier and parameters were used, which tools were available, and the expected output shape. A prompt that behaves differently under different settings is not one prompt.

  3. 3

    Stamp every generated output with the version identifier that produced it, and keep that stamp with the record so an output found later can be traced back to its exact instructions.

  4. 4

    Keep the active version per environment separate, so promoting a version in a test environment does not change what production runs. Publishing to production must be a deliberate act.

  5. 5

    Support rollback to any earlier version without discarding newer drafts. Rolling back must change which version is active, not delete the work that came after it.

Edge cases it handles

8
  • A published version must be immutable. Allowing an edit in place means every output stamped with that version is now attributed to text that no longer exists.
  • Model tier, parameters, available tools, and output schema must be recorded with the version. Without them, a reproduction attempt runs the same words under different conditions and gets different results for reasons nobody can see.
  • Rolling back must leave newer drafts intact and re-promotable, rather than treating rollback as a delete of everything after the target version.
  • Each environment must track its own active version, or a test promotion silently changes what customers receive.
  • Comparing versions is meaningless unless both are run against the same fixed evaluation cases. Store a shared set of cases and their results per version rather than judging on ad hoc examples.
  • An output stamped with a version that has been deleted must still resolve to something readable, so version records must be retained for as long as any output references them.
  • Version storage grows without limit if every keystroke is a version. Distinguish drafts, which may be overwritten, from published versions, which may not.
  • A run started against one version and completed after a promotion must be recorded under the version it actually used, not the one that is active when it finishes.

Definition of done

8
  • Publishing creates a new immutable version and never modifies an existing one.
  • Each version records the model tier, parameters, available tools, and expected output shape used with it.
  • Every generated output is stamped with the version that produced it and remains traceable later.
  • Each environment has its own active version and promotion between them is explicit.
  • Rollback changes the active version without deleting newer drafts.
  • Two versions can be compared against the same stored evaluation cases with results retained per version.
  • The feature matches the existing design system.
  • No existing functionality is broken.

Related features

How it works

  1. 1

    Copy the link

    Grab the Markdown instruction URL for this feature.

  2. 2

    Give it to your AI

    Paste it into Claude Code, Cursor, v0, Lovable — whatever you build with.

  3. 3

    It inspects, then implements

    Your agent reads your existing app first, then adds the feature to fit it.

Works with your stack

These instructions are written to adapt. They tell the agent to detect your framework, match your existing design system, and reuse what you already have — rather than assuming a particular stack.

Need it tighter than that? Customize the feature and tell it exactly what you're running.