Hey HN! At work, we've been experimenting with adding asynchronous agents into our development workflow. One challenge we've discovered is that agents have a really hard time following style rules in long-running tasks, no matter how much we beg or repeat these rules in our AGENTS.md.
To fix this, we've built Nudge. Nudge watches the code that Claude writes, and uses its Hooks API to prompt Claude with style rules whenever a rule violation is detected. This helps ensure rules that are loaded much earlier in the session (e.g. when reading AGENTS.md) are not forgotten later on during a long-running task or session. It's especially helpful for long lists of simple stylistic rules (e.g. "only add import statements at the top of the file") that are easy to forget.
To fix this, we've built Nudge. Nudge watches the code that Claude writes, and uses its Hooks API to prompt Claude with style rules whenever a rule violation is detected. This helps ensure rules that are loaded much earlier in the session (e.g. when reading AGENTS.md) are not forgotten later on during a long-running task or session. It's especially helpful for long lists of simple stylistic rules (e.g. "only add import statements at the top of the file") that are easy to forget.
You can see our full write-up at https://www.attune.inc/blog/nudge and you can use the tool today at https://github.com/attunehq/nudge. We hope this helps solve this pain point for other folks as well!