Back to directory

AI-DECLARATION.md

An open specification for a structured file that a software project includes to declare which parts of its code were generated or assisted by AI, and to what degree, so that a reviewer knows exactly which parts to double-check.

AI-DECLARATION.md

An AI-DECLARATION.md file uses YAML frontmatter for structured fields plus a required Notes section in the Markdown body for human context; at minimum it requires a version, an overall level, and the Notes section. A project can optionally break the declaration down further, giving individual levels to specific processes, such as design, implementation, testing or documentation, and to specific components identified by file path or directory, with the overall level required to be the highest level present among them and any process not listed implicitly assumed to be none.

The specification defines six levels describing the balance of human and AI action, running from none, where a human acts alone with no AI involvement, up through hint, assist, pair and copilot, to auto, where a human prompts and the AI brings the task to completion autonomously with no further involvement.

The project's stated goal is to make AI usage in a codebase a widespread, visible convention rather than to discourage using AI, so that a skeptical reviewer can look specifically at the declared parts. It is written in HTML for its specification site and released under the MIT license.

More in Code quality

See all 94
  • Critic

    critic.run

    A code-change viewer, backed by Y Combinator, built for reviewing work written by coding agents, presenting a plain-language narrative of each change alongside the diff and a running conversation with the agent.

  • GitIntel

    hoangsonww.github.io

    An MCP server that computes git repository analytics such as hotspots, churn, temporal coupling, knowledge maps and risk scoring entirely from local commit history, for AI coding agents like Claude Code and Codex.

  • Axint

    axint.ai

    A verification layer for AI coding agents working on Apple projects: it runs real Xcode builds and tests against the Swift an agent wrote, reconciles static analysis findings against that evidence, and returns a signed proof with the exact repairs still needed.