stack/agents/filbert/work/queue-150/r1/body.md at dev-console

0 Fork 1

Files

T

jason.woltje b9ea9304fd lead: commit filbert's untracked evidence (273 files: queue-101 queue-102 queue-105 queue-107 queue-108 queue-109 queue-110 queue-112 queue-113 queue-114 queue-115 queue-116 queue-119 queue-120 queue-122 queue-123 queue-125 queue-126 queue-128 queue-129 queue-130 queue-131 queue-132 queue-133 queue-134 queue-135 queue-136 queue-137 queue-139 queue-140 queue-141 queue-142 queue-145 queue-146 queue-147 queue-148 queue-149 queue-150 queue-151 queue-152 queue-153 queue-154 queue-155 queue-156 queue-162 queue-163 queue-165 queue-166 queue-167 queue-169 queue-170 queue-175 queue-177 queue-181 queue-182 queue-184 queue-185 queue-186 queue-187 queue-188 queue-49 queue-64 queue-66 queue-70 queue-71 queue-80 queue-84 queue-85 queue-87 queue-88)

Unclaimed files left in the canonical checkout. Inventory: canonical-cleanup-inventory.md (commit-as-evidence groups). detect-secrets scan clean on 2026-11-02: hits are sha256 pins, git object ids, test sentinels already in tracked tests and unit-file lines.

2026-11-02 15:05:14 -06:00

6.3 KiB

Raw Permalink Blame History

Row 150 round 1 review (Filbert): approve

Candidate: manifest sha256 17d118eb74ce887a1d0d37f044ed107de23c67072a9dc2203d704a01e35f9ec8, 7 paths, base eff8417b6, packet agents/darkwing/work/queue-150/ (request comment 28787). Brief: docs/plans/2026-10-15_talk-followups.md § "The human marks skipped messages read, and the Console says what talk prints" at blob b1c00f1285. Mos's decision 123 go covers public-diff.patch 096d922c75503db5bb67db0fb2e4c3b8be79d6d1ed9f1831ad81861ae3bfdca6 (sage-0560).

I found no defect. Every test the brief lists is there, and the broker change is a comment only. The notes below don't block. Note 1 needs a rebase at landing.

Rebuild

A fresh worktree at eff8417b6. r1/build.patch is a plain diff, so I applied it with git apply --index.
All 7 manifest hashes matched, and git diff --cached --check was clean (7 files, +204 -23).
The added lines have no dash, curly-quote or control characters.
The candidate's packages/webui/src/public/ diff has sha256 096d922c..., byte-identical to the public-diff.patch Mos approved.
The brief's after rows are in the base: row 142 is 02ee7f640 and row 146 is 277d1fa21.
Drift: one manifest path moved after the base. See note 1.

Brief tests, each found

All are in the new packages/cli/tests/mark-read.test.mjs unless named.

Marking one message. "inbox --read marks one message by the id inbox prints" marks a delivered message by its 8-character id without calling message.receive, then marks an undelivered one by its full id, and checks that the other messages stay unread.
An ambiguous prefix refused. The ambiguous-prefix test prints ambiguous, matches <id>, <id> with both full ids, sends no message.read for that argument, and still marks the next one. Prefixes shorter than 8 characters, including the empty string, refuse too-short.
An id that isn't the human's refused, with the rest still marked. The full id of a message the human sent to the coder goes to the broker as given and prints not marked <id>: message-not-found. A prefix with no match prints no-unread-match. The ids on either side are still marked, the output reads marked 2 of 4, and the command exits 2 with 2 of 4 not marked, which main writes to stderr.
--read-all leaves mosaic inbox at 0 unread. 230 messages, 100 delivered and 130 not, across two receive batches and two messages pages: it prints marked 230 of 230, then inbox shows messages: 0 unread, and a second --read-all prints marked 0 of 0.
The new Console text. webui send-browser.test asserts the new send-section sentence and the new could-not-read line in the real page.

What I checked and found correct

The not-recipient path. broker.mjs message.read fails message-not-found when the id isn't addressed to the human, before it checks delivery. So the receive-then-retry runs only for the human's own undelivered messages. A foreign or mistyped full id never drains the inbox.
The receive loop ends. message.receive returns the oldest 100 undelivered messages and delivers them. The loop stops when a batch holds the id or is empty, then retries the read once.
Which errors stop the command. perMessage treats a code as per-message only when busExit maps it to 2. outcome-unknown, invalid-response and the authentication codes (exit 1 or 3) stop at once, and the stop test pins that nothing is printed.
--read-all uses the walk inbox uses. unreadMessages is now the one walk for both. The targets are that list, oldest first, as the brief asks.
Flag combinations. --read with --read-all, with --json or with --tester, --read with no ids, and --read-all with ids all exit 4 before any call. tasks and agents refuse both flags. The combinations test pins all nine and checks that no call was made.
Docs. The cli README, TOOLS.md, the bus.js comment and the broker.mjs comment match the code and the test strings. The README states the side effect (a message that --read receives drops out of a later follow), and the first test pins it.

Notes (non-blocking)

Drift on broker.mjs. Row 139's landing (63c82f1c4) changed packages/bus/src/broker.mjs after base eff8417b6, so that path's manifest hash won't match on HEAD. The candidate's broker hunk is the comment above replies, and git apply --check of that hunk on HEAD's broker.mjs succeeds. The landing needs a rebase onto HEAD and the full gate on the new base, not on eff8417b6.
A second not-recipient is untested. No test reaches a not-recipient from the retry after the receive loop, as BUILD.md says. In the current broker I can't find a way to reach it for a human message. Every human delivery has holder_run null, which matches the human session, and a receive delivers any undelivered message to the human. If it ever happens, it is printed with its code like any other per-message refusal.
The brief lists not-recipient as a refusal to print. The candidate receives the message and retries first. I read that as within "the shape is Darkwing's call": without it, --read-all can't reach 0 unread when some messages were never delivered. The cost is that each receive also delivers up to 99 other messages, which then drop out of a later follow. Plain talk already leaves messages in that state, and the README says so. Sage may want this recorded as the chosen shape.
The per-message lines go to stdout. Only the summary N of M not marked goes to stderr. That meets "printed with its code". A script that reads only stderr won't see which ids were refused.

What I ran and what I took from receipts

I ran, on the slot Sage granted (sage-0571), from 2026-10-16T19:48:42Z to 19:49:21Z, as single files on the candidate with Docker disabled and TMPDIR set to scratch. All had 0 fail and 0 skipped:

Test file	Pass
cli mark-read	6
cli follow	6
cli cli	12
cli tester	7
cli app	30
cli format	3
webui send-browser	2

I ran no mutants, no suite and no gate.

From Darkwing's receipts, not rerun by me (BUILD.md "Checks", r1/gate-run.log): a targeted round (decision 136) from 19:22:58Z to 19:23:33Z that ran the same 7 files with the same counts and ended ROUND_PASSED. It ran no suite, no scripts/test-*.sh and no mutants. Sage's landing gate covers the suites, on the rebased candidate (note 1).

6.3 KiB Raw Permalink Blame History