I've noticed AI agents will run things like

gh pr checks 74 --watch --fail-fast 2>&1

. Unfortunately, this causes a complete waste of time when one of the PRs is not the root of the stack and the Github mergability check remains stuck on pending. This is annoying as a human too, since I sometimes have things notify me when ready.

Hi Bilal, thanks for flagging this. Which AI agent are you using? Most of them support a persistent instructions file, so you can add a rule that stops this behavior. Depending on the tool, that's `AGENTS.md` (used by most agents these days), `CLAUDE.md` for Claude Code, or `.github/``copilot-instructions.md` for Copilot. The rule would be something like: "Never run blocking watch commands such as gh pr checks --watch. Run gh pr checks <pr> once and report the current status. On stacked PRs not based on the default branch, expect mergeability to stay pending until the parent merges; don't wait on it." And if an agent is already stuck on a watch, most tools let you interrupt or cancel the running command from the session rather than waiting it out.

That's a workaround for agents, but it forces token waste on polling and doesn't address the root issue here. Even as a human, being able to watch for CI before running a follow-up command is useful. Can Graphite's CLI not offer their own equivalent here?

No, the Graphite CLI doesn't have an equivalent today. I've logged a feature request for one, and included your point that this is useful outside of agents too, for anyone who wants to be notified when CI is genuinely done on a stacked PR.

This is also quite annoying on Graphite's own UI. The checks bar shows some yellow portion as though things aren't done yet just because of the Graphite mergability check.

Noted, thanks for the feedback - i think this has come up internally as well!