Skip to content

backlog-burndown Ranks Every Open Issue, So a Round Can Dispatch a Worker at a Question Only the Maintainer Can Answer #1436

Closed

Description

@ptr727

Raised by a whole-unit pass run while reviewing PR #1435, outside the units that pass was scoped to, so it is filed rather than fixed there.

.agents/skills/backlog-burndown/SKILL.md step 1 says to rank every open issue, with no exclusion for the decision label. A parked question is by definition one only the maintainer can answer, so a round can rank one, group it, and dispatch a worker to an issue that has no fix in it.

The skill degrades rather than breaks: its step 5 already carries a "stopped on a question only the maintainer can answer" outcome, so the worker reports back rather than inventing an answer. What is wasted is the dispatch, the worktree, and the round the issue occupies in the ranking.

This gains weight with PR #1435, which makes the decision label the mechanism for enumerating the queue of questions waiting on the maintainer. Once that lands, the label reliably marks exactly the issues a burndown round cannot work, so excluding them is a one-clause change with a clear predicate rather than a judgment call.

Worth deciding at the same time whether a decision-labelled issue should be excluded from ranking outright, or ranked and reported to the maintainer as part of the round rather than dispatched, since a burndown round is one of the few moments that already has the maintainer's attention.

Activity

  1. ptr727 commented on Sep 8, 2026

    @ptr727
    OwnerAuthor

    A second, sharper instance of the same gap, from the whole-tree pass on PR #1435.

    .agents/skills/backlog-burndown/SKILL.md Run State says "a stopped group's question is a comment on its issue". So a burndown run parks its maintainer questions as comments on existing issues, which carry no decision label of their own.

    Once #1435 lands, the handoff enumerates this repository's open issues carrying that label and states the count. A question parked as a comment on an unlabelled issue is invisible to that enumeration, so the stated count omits every question a burndown run produced. That is the "queue nobody can see" failure the new rule exists to close, occurring in the fleet's highest-volume producer of maintainer questions.

    Two things to settle here alongside the ranking exclusion:

    1. Whether a burndown run labels the issue it comments a question on, so the enumeration reaches it, or files the question as its own issue instead.
    2. Whether a question parked as a comment is findable at all, since the enumeration is label-based and a comment carries no label independent of its issue.

    Also noted from the same pass, pre-existing and unrelated to #1435 except by subject: Run State says "The session ends, for a context limit or because the maintainer stops it". AGENTS.md "Session Scope" gives the end triggers as the branch changing, the pull request merging, or the next task being unrelated, and separately says compaction restarts context and the session keeps climbing rather than ending. #1435 removed the identical context-exhaustion claim from its own rationale for that reason, and this surface was not swept with it.

  2. ptr727 commented on Nov 1, 2026

    @ptr727
    OwnerAuthor

    Decision triage (2026-11-01) found the open question here already answered or overtaken, so it is not asked:

    Both questions are answered in the rule text, mostly by 06e45c4 (#3735). decision issues are excluded from ranking (backlog-burndown :150), an attended run asks them at the end of ranking (:146-149), an away run lists them and points to decision-triage (:786-787). An away park files its own decision issue or comments on an existing one (:693-697), an attended stop asks at once and writes the question on its issue (:421-428), and decision-triage labels an issue whose comment asks the maintainer something unanswered. One unrelated item remains as ordinary work: "Ending the Run" still says "The session ends, for a context limit" (backlog-burndown :810), against AGENTS.md "Session Scope".

  3. ptr727 commented on Nov 1, 2026

    @ptr727
    OwnerAuthor

    Closing per the maintainer's go-ahead in the backlog-burndown run of 2026-11-01. The decision-triage comment above already records both questions as answered in the rule text, mostly by #3735, which added away mode and retired unattended-handoff.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions