Developer logs AI agent rule violations to categorize and address recurring behavioral errors

On November 7, 2026, a developer writing in the AI section of DEV Community described an unusual practice: keeping a running log of every rule their AI coding agent broke. The log lives in plain text at the bottom of the AGENTS.md file, a convention many AI agents read at the start of every session. Each entry is dated and captures what the agent did and which working rule it stepped over, so the agent walks past its own record before it touches anything. The author says the log started as a way to make failures visible, but ongoing tracking delivered something more: a taxonomy. From the recorded breaches, the author distilled three recurring kinds of mistakes that AI agents make in real execution. The piece positions the log and the resulting classification as a practical tool for improving later interactions, since the agent re-reads its history each session and can carry the lessons forward. The material available so far consists of this single article. It does not state whether the log has continued to grow beyond the ten dated entries it mentions, and it does not name the three mistake categories explicitly, leaving both details open for later reporting.