How build receipts work

Evidence levels

• Self-reported

  Written by the builder in the submit form.

• Agent-drafted

  Drafted by the builder's AI agent from the build session, then approved by the builder. Still a claim, not a check.

• Agent-drafted + tool stats

  Agent-drafted, plus numbers gathered from local tooling (Git AI or AI co-author commits). Submitted by the builder, not verified by cobuilt.dev.

None of these levels is a verification. A future "measured" level, where cobuilt.dev reads attribution from the repository itself, will be the only one called verified.

How AI was involved

The builder chooses one level. None is better; honest is the point.

• Hand-coded

  Written without AI coding tools.

• AI-assisted

  I wrote the code; AI helped with completions, snippets, or reviews.

• AI-generated, human-reviewed

  AI wrote most of the code; I directed, reviewed, and edited it.

• Vibe-coded

  Built by prompting; I rarely read or edited the code myself.

What is sent, and what never is

Sent, after you approve the draft:

• Project name, description, links, type and tech stack
• The AI level you chose and the AI tools used
• Three short texts: what the AI did, what you did, what went wrong
• The agent's name and model, and optional summary numbers from Git AI or AI co-author commits

Never sent:

• Source code, diffs or file contents
• Prompts, chat transcripts or session logs
• Secrets, environment variables or local file paths

The server accepts only the fields above, with size limits. Tokens are stored as a hash and can be revoked in Settings at any time.