With the world model
replaying · step 001 · the readout, injected up front
vei oss attention-soon --repo nodejs/node CANDIDATE SET (unordered; listed by PR number): PR #65095 p_attention_24h=0.65 PR #65320 p_attention_24h=0.56 PR #65307 p_attention_24h=0.44 PR #65322 p_attention_24h=0.49 PR #65316 p_attention_24h=0.54 PR #65327 p_attention_24h=0.46 PR #65317 p_attention_24h=0.56 PR #65328 p_attention_24h=0.52 PR #65318 p_attention_24h=0.51 PR #65329 p_attention_24h=0.57 caveat: … the set is a starting point for your own inspection — not a verdict.
Ten candidates, no context. “Likely to be touched” and “a maintainer should act” are different claims, and only the second one is my job. First third of the budget: read all ten.
02 · view_pr × 10 — every lead
view_pr 65095 #65095 [open] ci=PENDING +1/-0 author=ulofiai(NONE) build: pass target architecture to small-icu genccode approved avivkeller (Aug 7), StefanStojanovic (today) · CI 75924 started 12:16 view_pr 65307 #65307 [open] ci=SUCCESS +5/-0 author=chasetonco(NONE) doc: add missing return types in fs.md five approvals · labels: author ready, commit-queue view_pr 65327 #65327 [open] ci=FAILURE@05:17 +613/-30 author=codebytere(MEMBER) fs: read small files in one thread pool round trip jasnell requested changes (AIX), then approved 06:33 — CI still red … 65316 65317 65318 65320 65322 65328 65329
Three shapes. 65307, 65317, 65320: green, approved, commit-queue — landing work. 65095: a new contributor's one-liner, twice approved, CI restarted two hours ago. And 65327: approved after a change request, but the AIX failure is still red — that approval doesn't make it safe.
03 · six sweeps of my own
list_prs --sort updated --limit 30 list_prs --sort stale --limit 25 search "author ready" --kind prs --limit 30 list_prs --sort updated --limit 30 --offset 40 list_prs --sort updated --limit 30 --offset 85 list_prs --sort updated --limit 30 --offset 130
Ten leads out of 794 is a sample, not a triage list — and they all have recent activity, so they're PRs a maintainer scrolling GitHub already sees. The rest of the budget goes hunting for work that is ready and stuck: recency pages past the fold, the stale end, and the label that means a contributor is done waiting on themselves.
04 · view_pr 65113 — a PR that is already in the tree
view_pr 65113 #65113 [open] +153/-7 author=unstubbable fs: fix `realpath` leaving symlinks unresolved after an unrelated stat approved by mcollina · landed as bcb680c0 — commit-queue bot never closed it mcollina pinged aduh95 on 2026-08-16 to investigate
This one has already landed. The commit is in the tree; the bot just failed to close the pull request. Pure housekeeping, invisible to every activity signal, ten seconds of a maintainer's time.
05 · view_pr 65276, 64512 — stuck on machinery
view_pr 65276 #65276 [open] ci=PENDING +31/-66 author=geeksilva97 sqlite: reuse cached column names in all() and get() approved trivikr · commit queue FAILED twice today, 06:43 and 11:02 view_pr 64512 #64512 [open] +20/-2 author=Nashit-h src: fix out-of-bounds write when transcoding odd-length ucs2 approved jasnell, Renegade334 · stuck since Aug 12 — one Jenkins node failed, every sibling job passed; contributor asked for a fresh CI
Same species: approved work waiting on a machine nobody has kicked. 64512 is an out-of-bounds-write fix — security-relevant, landable, held up six days by one unhappy node.
06 · view_pr × 8 more — the rest of the finds
view_pr 65346 stream abort fix · approved twice today · commit-queue set view_pr 65296 mcollina's ConnectionsList refactor · CQ failed 08:21, fresh CI 08:31 … 65159 61195 64801 62248 65196 63009
Eleven candidates of my own inspected. Some are real but not for today — a 231-day-old PR blocked on CI that won't start, a use-after-free fix that needs an actual reviewer. Noted, but a day that contains five ten-minute wins should spend itself there.
07 · merge the two pools — no more calls
Twenty-one candidates, ten slots. Five leads survive — the other five are waiting on CI, not on a person. Five of my own finds go in over them: the phantom-open PR, two commit-queue failures, the security fix, and a stream fix approved twice this morning. Rank by how little stands between each PR and being done.
Act now — ten picks, ranked
- #65307 merge five approvals, green, commit-queue — just land it
- #65346 merge approved twice today, green, commit-queue
- #65320 merge flaky-test fix, approved, in the queue
- #65113 close landed as bcb680c0; the bot never closed the PR
- #65296 merge re-queue once the 08:31 CI run is green
- #65317 merge routine dep bump, green, two approvals
- #64512 unblock OOB-write fix; one CI start unblocks it
- #65276 unblock read the CQ failure, tell the contributor
- #65095 merge confirm the fresh CI run, land the one-liner
- #65327 review approval predates the red AIX run — confirm first
Observed: CI states, approvals, queue failures, the landed commit — all as of the snapshot. Inferred, not established: that the readout's ten covered the day's real activity (I checked 21 of 794 PRs), and whether 65327's AIX failure is actually resolved — I recommended confirming, not landing.
Opus 5 + GitHub only
replaying · step 001 · survey the frontier
list_prs --state open --sort updated --limit 50 50 of 786 open PRs (sort: updated) #65346 +22/-0 opened today trivikr(MEMBER) [author ready …] … 49 more, the day's churn
The top of the list is whatever moved this morning. Somewhere in 786 open PRs is today's work; this shows me today's noise.
02 · try the other end
list_prs --state open --sort stale --limit 30
#51981 +30/-26175 76f opened 2024-03-06 (894d) darcyclarke(MEMBER)
Oldest first is mostly archaeology — an 894-day-old deletion PR. Staleness alone isn't a signal either.
03 · search the label that means "waiting on you"
search "author ready" --kind prs --limit 40 77 rows scanned PR #64133 [MERGED 2026-06-30] dgram: skip dns.lookup() for literal IPs …
Better idea, noisy result — the text search returns merged PRs too. The top hit closed six weeks ago. I'll work the open ones by hand.
04 · view_pr 65346
view_pr 65346 #65346 [open] ci=SUCCESS@05:22 +22/-0 trivikr(MEMBER) stream: reject pull on signal abort during flush [author ready, commit-queue]
Green, approved, in the queue. Keeper.
05 · view_pr 65296
view_pr 65296 #65296 [open] ci=FAILURE@Aug15 +71/-59 mcollina(MEMBER) http: use intrusive lists in ConnectionsList [author ready, c++] approved ShogunPanda, jasnell · commit-queue failed, fresh CI started today
Failed queue plus a fresh CI run — someone needs to re-queue this by hand.
06 · view_pr 65316
view_pr 65316 #65316 [open] ci=FAILURE +6/-11 nodejs-github-bot deps: update zlib to 1.3.2.1-motley-8002e91 [author ready, dependencies]
Bot dependency bump with approvals and a red run. Land once green — dependency updates shouldn't linger.
07 · view_pr 65276
view_pr 65276 #65276 [open] ci=PENDING +31/-66 geeksilva97(CONTRIBUTOR) sqlite: reuse cached column names in all() and get() approved trivikr · commit-queue failed twice today · author re-ran CI repeatedly
A contributor stuck in queue-failure limbo, re-running CI at a wall. This is exactly what "unblock" is for.
08 · view_pr 61195
view_pr 61195 #61195 [open] ci=FAILURE@2025-12-29 +4/-4 Ayoub-Mabrouk(CONTRIBUTOR) url: use ArrayPrototypePush in URLSearchParams methods
232 days old, last CI attempt in December. Real, but nothing suggests today is its day. Passing.
09 · view_pr 59685
view_pr 59685 #59685 [open] ci=FAILURE@12:42 today +27/-28 devholic22(NONE) test_runner: remove failure attribute in junit_report testcase 5 approvals · author inactive · stale-bot counting down CuriousStork (today): asked a collaborator to amend the message and land it
Five approvals and an auto-close timer. Someone asked, today, for a maintainer to finish it. Keeper.
10 · view_pr 65095
view_pr 65095 #65095 [open] ci=PENDING since Aug 7 +1/-0 ulofiai(NONE) build: pass target architecture to small-icu genccode [author ready]
A first-timer's one-line fix, approved, CI pending for ten days. Trivial and ready.
11 · view_pr 62943
view_pr 62943 #62943 [open] ci=FAILURE +173/-0 maruthang(NONE) test_runner: mock dual-package with conditional exports three approvals · open 114 days · CI re-triggered today
Three approvals and 114 days of waiting. Keeper, on fairness grounds as much as anything.
12 · view_pr 65207
view_pr 65207 #65207 [open] ci=SUCCESS +20/-7 leah-1ee(CONTRIBUTOR) doc: document per-architecture fast FFI argument limits
Docs, green, approved by the FFI owner. No reason to hold it.
13 · view_pr 65159
view_pr 65159 #65159 [open] ci=FAILURE +161/-49 jasnell(MEMBER) src: add v8::Local specialization for MaybeStackBuffer
A member's own C++ patch with red CI — he'll re-roll it himself. Not a maintainer-attention item. Passing.
14 · view_pr 64728
view_pr 64728 #64728 [open] ci=FAILURE +38/-47 trivikr(MEMBER) test: avoid timer race in event loop delay test [flaky-test]
Flaky-test fix that is itself failing CI. Ironic; passing.
15 · view_pr 57895
view_pr 57895 #57895 [open] ci=no-ci since 2025-04 +28/-18 ljharb(MEMBER) lib: make `new SafeSet(arr)` safe
No CI run in sixteen months. Dead end.
16 · page deeper into the recency list
list_prs --sort updated --limit 40 --offset 50
#53160 +566/-203 49f opened 2024-05-26 (813d) updated today szmarczak(ME…
Below the fold: old PRs that twitched today. A few worth opening.
17 · view_pr 65043
view_pr 65043 #65043 [open] ci=FAILURE +31/-3 mcollina(MEMBER) buffer: prevent string write offset overflow [author ready, buffer]
Overflow guard from a member, marked ready but red. Borderline; leaving it out of the ten.
18 · view_pr 65281
view_pr 65281 #65281 [open] ci=SUCCESS +30/-7 aduh95(CONTRIBUTOR) tools: fix the list of globals in ESLint config files five approvals · commit queue failed Aug 16, idle since
Five approvals, green, and a queue failure nobody re-ran. Keeper.
19 · view_pr 63552, 62983, 65113, 64697, 63457
view_pr 65113 #65113 [open] ci=SUCCESS +153/-7 unstubbable(NONE) fs: fix `realpath` leaving symlinks unresolved mcollina: already landed as bcb680c — bot failed to close; close it … 63552 repl crash fix (no CI yet) · 62983 needs-ci since May 64697 OOB ucs2 fix, red CI · 63457 crypto null checks, red CI
One gem in the tail: 65113 is already merged and just needs closing. The other four are needs-ci purgatory — real, not today's.
Act now — ten picks, ranked
- #65113 close already landed as bcb680c; bot failed to close
- #65346 merge green, approved twice, commit-queue
- #59685 unblock five approvals, stale-bot countdown, land it
- #65281 unblock five approvals, queue failed, needs a re-run
- #65276 unblock queue failed twice today, author is stuck
- #65296 merge confirm the fresh CI, re-queue
- #65316 merge bot zlib bump, land when green
- #65095 merge one-liner, approved, ten days pending
- #62943 merge three approvals, 114 days open
- #65207 merge green docs, owner-approved
Observed: everything above — labels, approvals, CI states, queue failures. Couldn't establish: which of the 786 PRs maintainers are actually moving on. I inspected 20; my picks are queue mechanics — approvals plus CI state — which finds what is landable, not necessarily what will be touched.