Your AI Coding Agent Is LYING When It Says "Done"

The article argues that repositories should define a strict, written definition of “done” for AI coding agents.

It proposes an AGENTS.md-based Task Completion Protocol that specifies the exact lint and test commands to run.

Agents must fix failures and report evidence before claiming completion, instead of relying on limited spot checks.

The goal is to prevent false success reports and broken handoffs to reviewers, CI, or later agents.