Repository navigation
π fix(ownership): issues the issue filter refuses reach the inbox as ADMIT - #16831
Conversation
β¦ADMIT A decided issue (approved-direction from /hive decision or /hive approve) and an agent finding where agents self-authorize sat in band "outside" with no verb when project.issue_filter.require_labels refused them: projectbluefin/testsuite#1091, server#622, dakota-iso#294 (decided) and testsuite#1085, documentation#1630 (L6 findings). Rule (9b) gives them ADMIT (decided_not_admitted, agent_finding_not_admitted) with a reason naming the labels that admit them. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com>
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: The full list of commands accepted by this bot can be found here. DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
Open PR overlap checkBase branch: Conflicts are checked with every PR merged onto the current tip of
|
β¦MIT too Doug's call on hivecommons#16831: an issue a person filed that only the missing project.issue_filter.require_labels keeps out gets ADMIT (person_issue_not_admitted) with "admit it so the hive works it, or close it". Holds, exempt, tracking, blocked, parks, reporter confirmation, reporter triage and standing dashboards are decided before the filter and keep their rows; a bot's issue stays outside. The ADMIT menu's reporter-trust note now shows only on untrusted_reporter. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com>
There was a problem hiding this comment.
The two rules the body describes are correct and well-grounded; the diff also ships a third rule the body says it does not, and that one changes inbox volume materially. I'd ask that the body and the code be made to agree before this lands.
Body/diff mismatch β person_issue_not_admitted. The What table's row "Person-filed, undecided" says outside β outside (unchanged), and the How section lists exactly two reason codes. But ownership.go adds OwnershipReasonPersonIssueNotAdmitted and the branch
if in.authorIsHuman {
return yourTurn(OwnershipReasonPersonIssueNotAdmitted, "A person filed this; the issue filter keeps it off the agent queue β ...")
}inside rule (9b), plus the authorIsHuman plumbing from RepoWorkIssue.AuthorIsHuman in decideRepoOwnership, the human_gate.go wording, the docs rows and the changelog fragment. The new test rows "a person's undecided issue" / "a person's undecided issue at L5" assert ADMIT. So on any repo with project.issue_filter.require_labels set, every open human-filed issue that does not yet carry the admission label becomes a your_turn=true ADMIT row β i.e. the whole untriaged backlog, not just the decided issues and agent findings the summary is about. That may be the intended product behaviour (it is symmetric with untrusted_reporter), but it is a different and much larger change than the one the body argues for, and the body currently tells a maintainer it does not happen. Either drop the third branch to match the stated scope, or update the summary/What table to own it and say why the backlog should surface as inbox rows.
What I verified and agree with:
- The match is exactly the filter's own refusal.
client.go:2040-2145decides exempt, tracking,blocked, reporter-confirmation, hard-suppress parks and reporter-triage beforeissueFilter.Admits, and only the last writes"Project issue filter";issue_admitter.gonow exports that string asIssueFilterReason, sobucketReason == github.IssueFilterReasoncannot pick up a park or exempt row. Matches the test rows that keepDECIDE/SPEC/not-yours. - Placement. Rule (9b) sits after the
TRIAGErow and before the safety net, so the L5 agent-finding case still yieldsTRIAGEas the table says. admitLabelsis read fromcfg.Project.IssueFilter.RequireLabelsForNotice()(config/issue_filter.go:173), no GitHub call;admitPhrasedegrades to "add an admission label" when empty.- Frontend.
label-filter.jsnow keys the reporter-trust note onown.reason_code === 'untrusted_reporter';reason_codeis the JSON tag ofOwnership.ReasonCode(ownership.go:177), so the new rows correctly skip that note. TestOwnershipFixedSets11565is bumped 32β35, consistent with three new codes (which is itself the tell that there are three, not two).
β hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
Hive status β updated in place; edits do not notify. Full history is in the Activity Monitor. Review findings from Read the diff against the v6 tree and traced the row it keys on; this looks correct to me.
CI was still in progress when I read this; the fixed-set count in |
β¦from main (#16843) * β¨ feat(ownership): CLOSE an agent finding whose cited lines are gone from main Follow-up to #16831. Review-backlog findings cite permalinks pinned to a commit (projectbluefin/testsuite#1085: SKILL.md:163). For a filter-refused agent finding the enumeration reads the cited lines at that SHA and the file on the default branch; when every citation's lines (>=16 chars, as in #16693) are gone, the row records RepoWorkIssue.CitedLinesGone and the dashboard offers CLOSE (cited_lines_gone). Bounded: 3 refreshes per repo per pass, 6h TTL, pinned-SHA content cached; read failures are unknown. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com> * π fix(ownership): stale-finding check settles unknowns and never CLOSEs a rename Hive review on #16843: - quality: a file gone from main is unknown, not gone (a rename keeps the lines); test pinned-present/main-404. - correctness: a pinned-SHA 404 or evidence too short is settled once and costs no further read or budget; `?query` is stripped from permalink paths (`SKILL.md?plain=1#L163`). - style: reuse hasAgentLaneLabel. - security: cache only the evidence lines (16-400 chars) at a pinned SHA, bounded at 1024 citations, never whole files. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com> --------- Signed-off-by: Danathar <Danathar@users.noreply.github.com> Co-authored-by: Danathar <Danathar@users.noreply.github.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: hivecommons-hive[bot] <280983584+hivecommons-hive[bot]@users.noreply.github.com>
Executive summary
Some issues that the issue filter (
project.issue_filter.require_labels) keeps off every agent queue showed up in no inbox, even though only a maintainer could move them. There were two kinds:issues_by_bandshowed them as bandoutside, with no verb andyour_turn=false, andreview_queueleft them out.TRIAGErow: that row needs the self-authorization hold, which is off at L6. So they also sat inoutsidewith no verb.This PR gives both kinds the existing
ADMITverb, with a reason that names the label that admits them. An undecided issue a person filed gets it too: the filter only ever lets it in through a maintainer's label.What
approved-direction(/hive decision,/hive approve)outside, no verbADMITdecided_not_admitted: "Decided, but the issue filter keeps it off the agent queue β addtriage/acceptedto admit it"outside, no verbADMITagent_finding_not_admitted: "β¦ addtriage/acceptedto admit it, or close it if it no longer applies"TRIAGETRIAGE(unchanged)outside, no verbADMITperson_issue_not_admitted: "A person filed this; β¦ addtriage/acceptedto admit it so the hive works it, or close it"outsideoutside(unchanged)Why
decideIssueOwnershiphad no rule for an issue whose enumeration bucket isfilteredwith reason "Project issue filter". Only the reporter-trust gate (untrusted_reporter) and the agentTRIAGErow produced an admission ask. A/hive decisionun-parks the issue and appliesapproved-direction, but not the filter's label. The issue then dropped out of every list while the reply told the maintainer it was waiting on them.How
pkg/dashboard/ownership.go: new rule (9b), placed afterTRIAGEand before the safety net. It applies to a row in bucketfilteredwith reasongithub.IssueFilterReason:approved-direction, it givesdecided_not_admitted;agent/<role>label, it givesagent_finding_not_admitted;RepoWorkIssue.AuthorIsHuman, now passed to ownership), it givesperson_issue_not_admitted.Only the filter's own refusal (bucket
filtered, reason "Project issue filter") counts. The enumeration records holds, exempt, tracking,blocked, parks (needs-specincluded), reporter confirmation, reporter triage and standing dashboards before it reaches the filter, with other buckets or reasons, so those rows keep what they had.Both use the verb
ADMIT. The reason namesproject.issue_filter.require_labels(ownershipEnv.admitLabels, read from configuration with no GitHub call).pkg/github/issue_admitter.go: exportsIssueFilterReason, so the dashboard matches the enumeration's own reason string.pkg/hiveadvisor/human_gate.go: plain-language wording for both codes;TestHiveAdviceWordsEveryOwnershipReasonkeeps the two lists in step.static-src/js/label-filter.js: the ADMIT menu note naming reporter trust's admission labels is now shown only onuntrusted_reporterrows. The filter rows' reason already names the filter's own labels, and those can differ.Files
src/pkg/dashboard/ownership.goadmitLabelssrc/pkg/github/issue_admitter.goIssueFilterReasonsrc/pkg/hiveadvisor/human_gate.gosrc/pkg/dashboard/static-src/js/label-filter.js,static/index.htmlsrc/docs/dashboard.md,changelog.d/β¦Tests
TestIssueFilterRefusedDecidedAndAgentFindingsAreAdmitTestRepoOwnershipAdmitsFilterRefusedRowsrequire_labelsread from configurationyour_turn_ui_11566_test.gogo testdashboard, hiveadvisor, adminmcp, writeback, githubnode --test src/pkg/dashboard/*.test.jsmake dashboard-gen-check,check-docs-citations.pyNotes
CLOSEwhen the cited lines no longer match main is the separate follow-up β¨ feat(ownership): CLOSE an agent finding whose cited lines are gone from mainΒ #16843. Ownership is decided without GitHub reads, so that check has to record a fact where the reads already happen.hive/claimed-by-β¦label. That is a separate item in the labels-vs-band checklist, and it doesn't change this row.origin/v6..origin/v5has no change toownership.go.π€ Generated with Claude Code
https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1