feat(queen): a runner bee that left no bundle is judged empty after the card's grace, behind TRIOS_QUEEN_NO_BUNDLE_EMPTY (Closes #1886)

[dmitrii-f-t27]

What this does

A runner bee that commits nothing leaves no bundle, so the reviewer's diff fails and the row answers wait for up to six hours while its issue stays claimed. This adds reviewer.t27's no_bundle_is_empty and wires it into the review sweep: with TRIOS_QUEEN_NO_BUNDLE_EMPTY=on, a finished runner row with no stored bundle is judged empty after a 10-minute grace, exactly as a container bee that committed nothing already is. A bundle that arrives inside the grace is still judged on its branch.

The flag is off by default: without it the code asks nothing new and every verdict is byte-identical to today's.

Provenance

This is a re-issue of #1891 by @gHashTag (original commit 216a6dc072), rebased onto today's queen so the dmitrii-f-t27 lane can carry it. The original commit did not apply cleanly: queen gained the region-merge check in the meantime. Both features are additive and are kept:

• queen-reviewer.ts and queen-tick.ts now carry both regionCheck (regions) and runnerBundleFacts (no-bundle);
• specs/PIN keeps today's holds hashes and the new reviewer hashes;
• queen-tick.ts gained the missing GC_RV_SEND_BACK / GC_RV_WAIT import that today's queen uses but does not import — without it bun run typecheck fails on queen itself.

The original author is credited in the commit trailer.

Verification

• biome ci . — 0 errors (120 pre-existing warnings).
• bun run --cwd apps/server typecheck — exit 0.
• bun test apps/server/tests/api/queen-no-bundle.test.ts apps/server/tests/api/queen-no-bundle-bench.test.ts — 17 pass, 0 fail, 1032 expect() calls.
• bun run --cwd apps/server test:api — 2220 pass, 72 skip, 18 fail; all 18 are environment-bound on this Mac (missing age-keygen, failing local git clone in vault/salvage fixtures), none touch the changed files. CI is the authority here.

Live context

/queen/status at 2026-10-18 02:59Z: capacity 70, active 12, unreviewed 328, reviewedLastHour 3, finishedLastHour 72. Measured 2026-10-11: 425 of 480 reviewer visits answered wait for a missing runner bundle. The six-to-seven-minute review rounds are also what makes the HUD print round stale.

Refs: gHashTag/t27#9021, gHashTag/t27#9069, #1886, #1891.

Dmitrii Fedorov (@dmitrii-f-t27)

[tri-claw]

Summary by t27-bees

Checked by the t27 compiler

_{t27c parse, parse-complete and typecheck on each changed .t27 file at c0e3103. Nothing was generated, built or run. A pass means the file is well formed, not that it is right.}

• trios/agent-server/specs/queen/reviewer.t27: parses and typechecks.

Read by the model

This PR adds a feature flag TRIOS_QUEEN_NO_BUNDLE_EMPTY (off by default) that makes the reviewer judge a finished runner row as empty when: (1) the diff cannot be read, (2) the runner stored no bundle for that issue, and (3) 10 minutes have passed since the turn ended. This matches how container bees that commit nothing are handled. The change wires a new runnerBundleFacts query and the no_bundle_is_empty Wasm function into the review sweep.

Worth a look

• trios/agent-server/apps/server/src/api/routes/queen-export.ts:337 — The runnerBundleFacts query checks EXISTS (SELECT 1 FROM queen_bundle b WHERE b.issue = d.issue) without filtering by conversation_id. If an issue has multiple attempts, a previous attempt's bundle will make a new attempt appear to have stored: true, preventing it from being judged empty even when its own bee committed nothing. The bundle check should likely include AND b.conversation_id = d.conversation_id (assuming queen_bundle has that column).

• trios/agent-server/apps/server/src/api/routes/queen-export.ts:351 — stored: row.stored !== false treats any non-false value (including null) as a stored bundle. The EXISTS subquery returns a boolean, so this is probably fine, but the defensive comment "an unreadable answer must not make finished work look empty" suggests concern about null; if the column is nullable, EXISTS still returns true/false.

• trios/agent-server/apps/server/src/api/services/queen-tick.ts:3256 — runnerLeftNoBundle returns false (continue waiting) when runnerBundleFacts returns null (no finished row for the conversation or unreadable finished_at). This is safe, but means a row whose dispatch record was cleaned up or corrupted will wait indefinitely rather than being judged.

• trios/agent-server/apps/server/src/api/services/queen-tick.ts:3455 — When runnerLeftNoBundle returns true, the code overwrites diff = { ok: true, files: [] }. This makes the review proceed as if the diff read succeeded with zero files, leading to an empty verdict. The logic is sound, but it silently converts a "diff unreadable" error into a successful empty diff — ensure this is the intended semantics (the tests confirm it is).

• trios/agent-server/specs/PIN:23 — The reviewer.t27 and reviewer.wasm hashes are updated. The new Wasm exports no_bundle_is_empty; the generated TS declarations in queen-reviewer-card.gen.ts include it. The PIN update looks correct and matches the compiler output shown.

---

_{Read: 11 files, +1394 -6, the whole diff up to 60000 characters (the rest was not read). Head c0e3103. Model: openai-compatible/nvidia/nemotron-3-ultra-550b-a55b. Commands: @t27-bees review | summary | help | pause | resume. Free for every repository.}

[github-actions]

✅ Tests passed — 3915/3988

Suite	Passed	Failed	Skipped
✅ agent	429/429	0	0
✅ build	9/9	0	0
✅ cdp-protocol	5/5	0	0
✅ eval	93/93	0	0
✅ server-agent	280/280	0	0
✅ server-api	2273/2345	0	72
✅ server-browser	6/6	0	0
✅ server-integration	10/11	0	1
✅ server-lib	279/279	0	0
✅ server-pglive	166/166	0	0
✅ server-root	68/68	0	0
✅ server-sim	8/8	0	0
✅ server-skills	31/31	0	0
✅ server-tools	244/244	0	0
✅ shared	14/14	0	0

View workflow run