Bugfix — Issue #1073: fix-bug factory triage mis-selects unlabeled enhancement as a bug

Problem

The autonomous bug-fixing factory (fix-bug.yml, rulebook ai/FIX-ISSUE.md, triage in ai/AGENTS.md #21) is supposed to pick the newest open bug to fix. Instead, on this run it picked issue #1073 "Enhancement: Add Metrics & Trends dashboards…" as its bug candidate.

Why: fix-bug.yml selects "the newest open issue that does NOT carry a work-type label (enhancement, task, investigation, documentation)". A batch of 21 issues (1053–1073) was opened on 2026-09-06 without any label. Those issues are all feature/analysis proposals (enhancement feature-requests plus #1054 ISTQB Compliance Review), but because they carried no work-type label the factory cannot tell them apart from genuine bugs — so it kept returning the newest one (#1073) and burning an entire 2-hour agent run trying to "reproduce" a feature request. If left unlabeled, the next 21 fix-bug runs would each waste their budget on one enhancement.

Measured evidence (pre-fix)

• Open issue count: 190.
• Label histogram of all open issues: task → 158, enhancement → 11, unlabeled [] → 21, bug → 0.
• Triage command output (pre-fix):

  gh issue list --state open --limit 200 --json number,title,labels,createdAt \
    --jq 'map(select([.labels[].name] | any(. == "enhancement" or . == "task" or . == "investigation" or . == "documentation") | not)) | first'
  # => {"createdAt":"2026-09-06T10:58:15Z","labels":[],"number":1073,"title":"Enhancement: Add Metrics \u0026 Trends dashboards..."}
  

• #1073 body verified by reading it: a feature proposal ("Add a dedicated Metrics & Trends area under the Reports menu…") with an Acceptance-criteria list. No symptom, no regression. Not a bug.

Root Cause

1. ai/AGENTS.md #21 defines triage: fix-bug.yml picks the newest open issue without a work-type label.
2. Issues 1053–1073 were created with labels: [] (metadata defect — the creators did not apply the required work-type labels).
3. Because those issues are invisible to the factory's skip-list, they are indistinguishable from bugs, so the triage returns them as "the bug to fix".

Blast radius: 21 unlabeled issues out of 190 (≈11%). No application code path affected — the defect is purely in the CI triage metadata.

Fix (minimal)

Per AGENTS.md #21 ("ALWAYS label them correctly so the triage works"), each unlabeled issue got its correct work-type label:

• 20 feature-request issues → enhancement: gh issue edit 1053 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 --add-label enhancement
• #1054 "ISTQB Compliance Review" (analysis/study → matrix + backlog, not a feature) → investigation: gh issue edit 1054 --add-label investigation

Verification (error-free)

1. Triage now returns no bug candidate: the jq from above emits empty.
2. Bug-label open count: 0 (no genuine open bug ever existed).
3. No unlabeled open issues remain: 0.
4. Labels read back: #1073 enhancement, #1054 investigation, #1068 enhancement.

Regression

tmp/TLU_Test_Cases.md → Regression — Issue #1073 suite (RF-1073.1 … RF-1073.7), result 7/7 PASS. No application code or i18n bundles touched.

Files changed

None (metadata-only fix on GitHub issues; no repo source files altered on this run — this document and the TLU suite are the only local artifacts).

Refs

#1073 — this issue was the selected target; the actual defect fixed is the factory triage metadata that caused enhancement #1073 to be mis-selected as a bug.