They're Poisoning the Agents!

Mitchell Hashimoto is correct to poison his AI agent files with prompt injections that expose unreviewed submissions to open-source projects.

• Agent file poisoning — Instructions were added to the Ghosty agent's markdown file directing any PR creation to include a confession of being a sad dumb AI driver with no real skills [00:31]

• Instant ban outcome — Users caught submitting the injected code receive immediate bans from the project [00:44]

• Time disrespect problem — Submitters who skip review expect maintainers to read, integrate, and support code they never examined themselves [05:09]

• tldraw policy shift — The project began automatically closing external PRs without a vouch after months of low-quality AI contributions [03:02]

• Future career impact — Banned contributors risk blank rejections on job applications when teams check public contribution histories [06:02]

• What phrase do poisoned AI agents add to pull requests?

• How can running unvetted AI agents lead to committing secrets to GitHub?