Skip to content

Auto-skip when a triager has already left a similar comment #76

Closed

Description

@potiuk

When a triager (any user with the triager role, including the bot itself) has already left a comment on a PR and the classifier would now propose a similar comment, classify the action as skip automatically rather than queuing the near-duplicate for human confirmation.

Why: today the human triager has to repeatedly dismiss "we already said this" proposals. It wastes attention and risks accidental duplicate posts when the operator clicks through too quickly.

Open questions

  • Similarity definition: exact rule-class match (same template / classifier rule fired before), keyword overlap, or embedding similarity? Rule-class match is the simplest and probably catches the bulk of duplicates.
  • Threshold knob exposed to the user via config?
  • Logging: silent skip, or a brief note in the progress output ("auto-skipped — same as existing triager comment by @user")?

Related but distinct: Issue for "repeat-triage avoidance when maintainer already engaged" — that is about any engagement; this one is about similarity of the proposed comment.

Activity

  1. added
    mode:TriageAgentic Triage — spot, classify, route, surface duplicates
    kind:policyRule changes (eligibility, thresholds, behaviour switches)
    on May 7, 2026
  2. potiuk commented on Aug 5, 2026

    @potiuk
    MemberAuthor

    Status check against main — the core ask is implemented; one specific gap remains.

    Done. Rows 3–4 of the decision table (classify-and-act.md:83-84) auto-classify as skip when a triage marker is present after the last commit — already_triaged → skip, no prompt. The marker is either a viewer comment carrying the Pull Request quality criteria link or the pr-triage-fold body block with a matching head SHA. Row 0 covers the stale first-time-contributor case.

    It uses the option this issue itself preferred — "Rule-class match is the simplest and probably catches the bulk of duplicates" — which also makes the proposed threshold knob moot: a marker either exists or it does not, so there is nothing to tune. That open question can be closed.

    The gap is the word "any". This issue says any user with the triager role; the implementation is viewer-scoped — the field is literally viewer_triage_fold_present, and row 3 reads "Viewer triage marker present". If another maintainer triaged a PR under their own account, your session sees no marker and re-proposes the same comment.

    That is the same harm described here — wasted attention, risk of duplicate posts — just between two people rather than inside one session. With new committers about to start triaging, it stops being hypothetical.

    Remaining work, narrowed: make marker detection recognise a triage marker left by any triager, not only the viewer. The pr-triage-fold body block is the natural carrier since it is a single shared block on the PR body rather than a per-author comment; the comment channel would need the marker match widened beyond the viewer's own comments.

    The logging open question is also still open: rows 3–4 report "Already triaged M days ago" without naming who triaged it, which matters once it can be someone else.

  3. Kaap10 commented on Nov 15, 2026

    @Kaap10
    Contributor

    Multi-triager marker recognition and triage comment attribution were merged in PR #1420 (commit 7eb89ad). Closing this issue as resolved.

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

    capability:triageSweep + classify + propose dispositionfamily:pr-managementpr-management-* skillskind:policyRule changes (eligibility, thresholds, behaviour switches)mode:TriageAgentic Triage — spot, classify, route, surface duplicates

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions