[FEATURE] Change automated duplicate detection from auto-close to advisory mode

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

The current automated duplicate detection system auto-closes issues after 3 days unless users actively contest by adding a 👎 or comment. This creates a problematic dynamic:

1. Users cannot verify "same root cause" — The codebase is closed-source. When the bot suggests an issue might be a duplicate, users have no way to verify whether the underlying cause is actually the same. We can only observe symptoms, not the code.

1. Burden of proof falls on users — Users must actively defend their issues against a judgment they cannot verify. If they don't respond in time (vacation, busy week, timezone differences), valid issues get closed.

1. The system makes mistakes — Issues have been closed as duplicates of each other in circular chains (#19267 documents 32 such cases). Issues have been suggested as duplicates of already-closed issues (#20282).

The structural contradiction:

• GitHub Issues says "we want your feedback" (open)
• Closed-source means "you can't see the code" (information asymmetry)
• Auto-close says "prove it's not a duplicate" (burden on the uninformed party)

Proposed Solution

Change the automated system from auto-close mode to advisory mode:

| Current | Proposed |
|---------|----------|
| Bot suggests duplicates | ✅ Keep (helpful) |
| Bot adds labels for triage | ✅ Keep (helpful) |
| Bot auto-closes after 3 days | ❌ Remove |
| Human maintainer reviews and closes | ✅ Add |

The bot should assist maintainers by surfacing potential duplicates, not replace maintainer judgment on whether issues are truly the same.

Alternative Solutions

1. Keep auto-close but require higher confidence — Only auto-close when the bot is highly confident (e.g., near-identical reproduction steps). Current threshold seems too low.

1. Keep auto-close but extend the window — 3 days is short. 14 days would give users more time to respond.

1. Keep auto-close but exclude closed issues from suggestions — At minimum, don't suggest already-closed issues as duplicates (#20282).

I believe the proposed solution (advisory mode) is best because it addresses the root issue: AI should augment human judgment, not replace it — especially when users lack the information needed to contest AI decisions.

Priority

Medium - Would be very helpful

Feature Category

Other

Use Case Example

1. I report a bug: "Ghost character ) appears in output"
2. The bot suggests it might be a duplicate of a statusline-related issue
3. I look at the suggested issue — the symptoms are different () vs c), but I can't see the code to know if the root cause is the same
4. Current: I must guess whether to contest or accept, under threat of auto-closure
5. Proposed: The bot labels it for maintainer review. A maintainer who can see the code decides whether it's truly the same issue

Additional Context

I genuinely support Anthropic's direction of using AI to reduce maintainer burden. I want this experiment to succeed. This feedback comes from a place of wanting the system to work better, not opposition to AI-assisted issue management.

Related issues:

• #19267 — Documents 32 circular duplicate chains
• #20282 — Bot suggesting already-closed issues as duplicates

---

✍️ Author: Claude Code (DevContainer) with @carrotRakko

Note: This issue was written and submitted by an AI agent (Claude Code), with human review and approval.