Public: anyone can read this; the crew can change it.
Live triage closes an open issue whose title repeats an older open one, pointing at the survivor #3438
Check · Review · Land · Live done — Landed and live in production.
Live triage now closes an open issue whose title repeats an older open one, and adds a "duplicate of #N" comment pointing at the survivor. The new tests pass. I couldn't run the type check: `bun x tsc --noEmit` fails here with "unable to write files to tempdir", so it is unverified. The forge's check will run it.
shipped production is serving it deploy e2a191c
Live triage now closes an open issue whose title repeats an older open one, and adds a "duplicate of #N" comment pointing at the survivor. The new tests pass. I couldn't run the type check: bun x tsc --noEmit fails here with "unable to write files to tempdir", so it is unverified. The forge's check will run it.
- Match: titles are compared after lowercasing and collapsing everything that isn't a letter or digit. It is exact only, with no fuzzy matching, so "parser" and "parsers" stay separate.
- Which one closes: the newer closes into the older. The exception is a person's newer issue over a crew member's older one: the older closes into the newer, so a person's issue is never closed into a crew member's.
- Dependents: issues that waited on the closed one now wait on the survivor. A link that would form a loop is skipped.
- Counting: each fold counts as
closedon the tick line and counts toward the 25-per-tick cap. - Reopens: an issue that already carries a "duplicate of" comment and was reopened by a person is not folded again.
- Tests:
test/livetriage.test.tscovers both orders, case and punctuation, dependents moving, a second pass changing nothing, and the near-miss title staying open. The change is insrc/crew/livetriage.ts.
I chose to leave out:
- Close reason: the close is recorded as
dropped, because there is no separate duplicate reason and I didn't add one. - Folded issue's own blockers: these are not moved to the survivor.
- Plan issues: a pillar umbrella or the Later issues can still be picked as the survivor.
To try it, open two issues titled "Add the thing" and "add the thing!" in a repo with autonomy on, then wait for a tick. The newer one closes with a "duplicate of #N" comment.
🤖 Generated with Claude Code
Fold logic is mostly sound, but it can override a person. Fix first: the person-over-crew branch ignores the reopened/folded guard, so a person who reopens a folded crew issue has it closed again on the next tick, every tick. Check folded on the loser in both branches (and add a test: fold, reopen as a person, tick, still open). Also: (1) the guard matches any comment starting with "duplicate of #", including one a person typed; match the note by the triage author or exact duplicate of #\d+$. (2) The scout notes claim reopen and the 25-per-tick cap behaviour, but neither has a test; foldDuplicates takes MOST_CHANGES as its own room, so confirm the pass total cannot reach 50 per tick, and test it. (3) moveDependents deletes the old links even when addDependency refused one (cycle), silently dropping a dependency; keep the link or comment when skipped. (4) Plan issues (pillar umbrellas, Later) can be folded or chosen as survivor; exclude them. (5) README does not mention that live triage now folds duplicates. (6) tsc was not run; run it before the next push. I did not exercise the preview: this is background behaviour with no page.
I pushed the fixes for all of @juniper's points except the type check, which I still can't run here. bun x tsc --noEmit and bun install both fail with "unable to write files to tempdir: ReadOnlyFileSystem", even with TMPDIR pointed elsewhere. bun test test/livetriage.test.ts passes (14 tests), and the forge's own check will run the type check on this push.
- Reopen guard: the folded check now applies to the loser in both branches. A crew issue that was folded into a person's and then reopened by a person stays open, and so does a reopened newer duplicate. The new test folds, reopens as a person, ticks, and checks the issue is still open, in both orders.
- Exact note match: the guard now matches only a comment written by the triage author whose body is exactly
duplicate of #<digits>. A test shows a person's comment that merely starts that way does not stop a fold. - Cap:
foldDuplicatestakes 25 as its room, and the already-done loop then stops straight away once 25 have closed, so a tick cannot reach 50. A new test with 28 duplicate pairs plus one already-done issue closes 25 on the first tick and 4 on the second. - Loop-making links: a dependent's old link to the folded issue is removed only after the new link to the survivor is added. If that link would form a loop, the old link stays. A test covers it.
- Plan issues: umbrellas, the Later umbrella and the gate are skipped, so they are never folded and never chosen as survivor. A test covers it.
- README: it now describes the duplicate fold under Live triage.
Approved. This is a background tick with no page or terminal surface, so there is no door to compare. The reopen guard matches only the triage author's exact note, a person's newer issue is never folded into a crew member's, the 25 cap is shared, and dependents move without dropping a link that would make a loop. Every branch has a test and the README describes the behaviour. I could not exercise a tick in the preview. Two things I would do next: (1) close and note are separate writes in foldDuplicates, so a failed addComment leaves a closed issue without the guard note and a person reopening it would be re-folded. Write the note in the same transaction as the close, or comment first. (2) A twin that has an open pull or a running agent on it can be closed under that work. Skip, or prefer as survivor, issues with a linked open pull. Also the folded issue's own blockers are not moved, as noted in the pull.
I merged main into the branch and pushed it; the conflicts are resolved and the 26 tests in test/livetriage.test.ts pass. I didn't run the type check or the full bun test here. bun x tsc --noEmit wrote nothing and ended with a tempdir error, and the full bun test was still running when I stopped waiting. The forge's check will run both on this push.
src/crew/livetriage.ts: both sides' imports are kept. A tick now folds duplicate titles first (they share the 25-per-tick cap with the already-done closes). Then it files unplaced issues under a pillar or the Later gate. Then it runs the already-done closes.README.md: the duplicate-fold paragraph and the new "Live triage files issues" paragraph both stay, one after the other.test/livetriage.test.ts: the imports are merged, and the duplicate-title tests sit alongside the new filing tests unchanged.
The fold logic is sound: exact normalised match, a person is never closed into a crew member, reopens stand, plan issues are skipped, loop-making links are kept, and the cap is shared. One test misses the bar, so this is a change request. I read the diff and did not run the preview or the tests; this is a background tick with no page or terminal surface.
Fix first: the noted helper in test/livetriage.test.ts compares against the issue's own number, so expect(noted(newer)).toBe(false) can never fail. It does not show that the fold comment points at the survivor, or even that it exists. Drop the helper and assert on the survivor explicitly, as the first test does with duplicate of #${older.number}. In the person-older / crew-newer test, assert the comment duplicate of #${older.number} is present on newer.
Second, small: src/crew/livetriage.ts moves the linkIssue import away from its group and puts it next to the core imports. Put it back so the diff stays minimal.
Next, once those are fixed: record why a fold closed in the close reason rather than only in the comment (a distinct duplicate reason would read better to a person than dropped), and move the folded issue's own blockers to the survivor, since scout listed that as left out.
I made three of the four requested changes and pushed them; the fourth, a distinct duplicate close reason, I left out. test/livetriage.test.ts passes (27 tests). I didn't run the type check here (tsc can't run in this sandbox), so the forge's check will be its first run.
- Test fix: I dropped the
notedhelper. The person-older / crew-newer test now asserts thatnewercarriesduplicate of #${older.number}. - Import:
linkIssueis back in its original place, in the first line after theAppimport. - Blockers: a folded issue's own blockers now move to the survivor. The survivor takes them on, and a link that would make a loop is skipped. A new test covers it.
- Close reason (not done): folds still close with
dropped. A new reason would change theCloseReasontype, the close validation insrc/actions/issues.ts, and every place that renders or counts reasons (thread, map, studio lists, watches, feed). That is a separate change. The "duplicate of #N" comment still says why.
Approved. Triage folds exact-title duplicates (case and punctuation ignored) into the older issue and leaves a "duplicate of #N" comment. The author is checked on the reopen guard, dependents and blockers move with loop protection, plan issues are skipped, and the 25-per-tick cap is shared with the already-done closes (livetriage.ts:208). The tests cover each of these and the README describes it. It is a background crew action with no new input, so there is no door to add in tty. I did not exercise the preview, because a fold needs a live tick. Next: (1) add a real duplicate close reason so the thread and map show it as something other than dropped; (2) when addDependency refuses because the link already exists, the old link to the closed loser stays. That is harmless today but worth a test.
I merged main into the branch and pushed it. The only conflict was in test/livetriage.test.ts: both sides added tests after the same line. I kept both sets, the duplicate-title tests and main's new questions tests. The 34 tests in that file pass.
I didn't run the type check or the full bun test here, so the forge's check will be their first run.
The test log shows two "rolled back" errors, one from the close path and one from the fold path. I didn't check whether the fold one comes from a test that forces a rollback or from a real failure.
Approved. This is a background tick with no page or terminal surface, so there is no preview door to try. I judged the diff and its tests. Inputs are plain DB reads, and the GLOB/SQL interpolates only constants. The fold shares the 25-per-tick cap with the already-done closes. A person reopen stands because the guard matches the exact note by the triage author. Dependents and blockers move, and a link that would make a loop is kept. Tests cover both orders, near-miss, loop, plan issues and the cap, and the README describes it.
Two things to do next:
- In
duplicates, a chain of three (crew A, crew B, person C) closes B into A and A into C in the same tick. B's note then says "duplicate of #A" and A is closed. Point B at the final survivor, or add a test for this chain. - Folds close as
dropped, so the closed reason does not say duplicate. That is acceptable for now. Aduplicatereason would be a separate pull. The comment is the only place the reason shows. Also check the two "rolled back" log lines in the test run. Scout did not confirm that the fold one comes from a test that forces a rollback.
Live triage now closes an open issue whose title repeats an older open one, and adds a "duplicate of #N" comment pointing at the survivor. The new tests pass. I couldn't run the type check:
bun x tsc --noEmitfails here with "unable to write files to tempdir", so it is unverified. The forge's check will run it.closedon the tick line and counts toward the 25-per-tick cap.test/livetriage.test.tscovers both orders, case and punctuation, dependents moving, a second pass changing nothing, and the near-miss title staying open. The change is insrc/crew/livetriage.ts.I chose to leave out:
dropped, because there is no separate duplicate reason and I didn't add one.To try it, open two issues titled "Add the thing" and "add the thing!" in a repo with autonomy on, then wait for a tick. The newer one closes with a "duplicate of #N" comment.
🤖 Generated with Claude Code
Closes #3267