Skip to content

Use case: AI code provenance #30

Closed

Description

@adityamukho

Summary

Add a use case: record which code an AI agent wrote, under which prompt and model, and every human edit after.

The idea

When an agent writes or changes code, store facts about the change: the function or line range, the agent, the model, the prompt or task, and the commit. Later human edits are new facts on the same code. History is never lost.

Questions it answers

  • Which code in this release is still unedited agent output?
  • Which bugs trace back to code written under a given prompt or model?
  • Who (or what) last changed this function, and what did it look like before?

Why Minigraf fits

Suggested content

  • The idea and why teams will ask these questions
  • A fact schema (code unit, authored-by, model, prompt, edited-by, commit, bug)
  • A query for unedited agent output in a release
  • A query tracing a bug back to the authoring prompt

Notes


Generated by Claude Code

Activity

  1. adityamukho commented on Oct 19, 2026

    @adityamukho
    ContributorAuthor

    Builds on the git sidecar idea in #34, which provides the commit and entity index this use case attaches facts to.


    Generated by Claude Code

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    documentationImprovements or additions to documentation

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions