You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Phase 6.4 — My pull request is my turn when someone leaves feedback as a comment since my last push #373
At work, colleagues often answer a pull request of mine with a comment on the pull request rather than a review, and nothing tells me the next move is mine. Since #371, a review from someone else after my last push makes my pull request my turn, but feedback written only as a top-level comment does not, so it is missed. For my own pull requests the fetch reads the author and time of the newest 100 top-level comments, enough to tell whether I replied, but not which comments ask something of me.
Ideal state
A pull request I opened is my turn when someone else left a top-level comment on it after my last push and I have not replied since
A comment from a reviewer bot, such as Copilot, makes it my turn like anyone else's
A comment from a bot that only reports, such as a CI, coverage or preview bot, does not
It holds however many comments the pull request has
The turn ends when I push to it, comment on it, submit a review, or it is merged or closed
Out of scope
Comments on other people's pull requests. My review of someone else's pull request already takes their comments into account, and comments added after I reviewed are for the author to act on
How does hub.toml say which bots' comments count: a list of bots to react to, or a list of bots to ignore? (a file format)
Starting points
domain/src/pr_turn.rs — TurnReading::of, which has no reason for a comment today
clients/src/github/prs/fetch.rs — the pull request query, which takes the first 50 top-level comments with their bodies, and the newest 100 by author and time for my own pull requests
#195 plans to hard-code a list of known bots so the PR detail view can collapse their comments. This issue's open decision, whether hub.toml names bots to react to or bots to ignore, should settle one classification of bots that #195 uses too, so hub does not end up with two lists that disagree.
Since #371 (c823c4e), the query for my own pull requests also reads the newest 100 top-level comments, by author and time only, to tell whether I replied to a review. The newest comment is always among them, however long the conversation. The first 50 comments with their bodies are still what the investigation prompt gets.
#371 already decides bots one way: a reviewer whose GitHub account type is Bot counts like anyone else when its review has inline comments, and its review with none, such as Copilot's overview, asks nothing (domain/src/review_waiting.rs). The open decision here should settle whether comments follow the same rule or a configured list, so hub does not end up with bot rules that disagree.
Activity
ooloth commentedon Oct 19, 2026
#195 plans to hard-code a list of known bots so the PR detail view can collapse their comments. This issue's open decision, whether
hub.tomlnames bots to react to or bots to ignore, should settle one classification of bots that #195 uses too, so hub does not end up with two lists that disagree.ooloth commentedon Oct 19, 2026
Since #371 (c823c4e), the query for my own pull requests also reads the newest 100 top-level comments, by author and time only, to tell whether I replied to a review. The newest comment is always among them, however long the conversation. The first 50 comments with their bodies are still what the investigation prompt gets.
ooloth commentedon Oct 19, 2026
#371 already decides bots one way: a reviewer whose GitHub account type is
Botcounts like anyone else when its review has inline comments, and its review with none, such as Copilot's overview, asks nothing (domain/src/review_waiting.rs). The open decision here should settle whether comments follow the same rule or a configured list, so hub does not end up with bot rules that disagree.