flipback plugin uses Stop hook to catch unfinished Claude Code turns

On November 12, 2026 at 19:06, the AI section of DEV Community published an article introducing flipback, a Claude Code plugin. The plugin reads the transcript when a turn ends and checks whether the user's request still contains clear unfinished steps. If it finds any, it prompts the agent to keep going rather than closing out the turn. The article was generated with AI from flipback's published code, replay results, and offline tests, and it also describes practices such as using /goal to set an verifiable endpoint before starting a task.

The article walks through a typical case: you ask a coding agent to do three things at once, for example add a slugify() helper to src/text.ts, write tests for it, and update the README with an example. Instead of finishing all three, the agent adds slugify() to src/text.ts, explains that it lowercases, strips punctuation, and collapses whitespace into hyphens, and then says it will now write the tests and update the README, with the turn ending right there. The piece notes that nothing in that message is false, but the tests and the README example have not actually been done. That is exactly the situation flipback is meant to catch, flagging the unfinished work so the agent resumes instead of stopping prematurely.