stack/agents/filbert/work/queue-154/r2/body.md at proxy-ui

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

5.9 KiB

Raw Permalink Blame History

Filbert review, row 154 round 2: changes

Candidate manifest 1db8d410ca4d706a355ce3d1dec3e985ae058093a4a56f5a31521079a4cce017 (195 lines), r2/candidate.patch (3c7c0114...) applied with --index to 850484a7d. The manifest matched before and after every run.

One change is requested: the candidate fails its own provenance test. My r1 change (a fallback test) is addressed, and the fixes for Dewey's r1 findings hold up. Deviations (i) to (iii) are Dewey's to rule on; I give no verdict on them.

Change: `editor/BUILD.json` records an `editor.js` that is not the shipped bundle, and `editor-provenance.test.mjs` fails

The candidate's packages/webui/editor/BUILD.json records src/public/editor.js as 73975eb2cdc4262c6b5488829ad5349e8ad1358ad6631fb51bc334d6c6bbbb80, 758168 B. The shipped src/public/editor.js is 32fcaa18eaee2ff426c3c7e06038ee190f9fd7a332c9f3db349aa74df22c9fb5, 758277 B. Manifest line 14 and BUILD.md (:400, :428) give that same value. The hash 73975eb2 appears nowhere in the packet's .md files.

My rebuild with the pinned toolchain (lock eb60a494..., esbuild 0.28.2 binary e1698a3d..., toolchain check rc=0 under MOSAIC_REQUIRE_EDITOR_BUILD=1) gives:

editor.js 32fcaa18..., 758277 B, the same as the shipped file;
editor.css 8654ac57..., 4462 B, the same as the shipped file;
a BUILD.json that differs from the committed one only in that editor.js entry (lines 7-8).

So the bundle is reproducible, but the record of it is stale. With the candidate as shipped, under MOSAIC_REQUIRE_EDITOR_BUILD=1, editor-provenance.test.mjs fails 4/6:

:79 "the bundle is the recorded build" fails at :89: src/public/editor.js is not the recorded build, actual 32fcaa18..., expected 73975eb2....
:159 "rebuilding from the source gives the same bytes" fails: src/public/editor.js rebuilt, actual 32fcaa18..., expected 73975eb2....

That conflicts with BUILD.md :238, :405 and :430 (editor-provenance 6/6, webui 201/201). The patch may have been cut from a tree whose BUILD.json was not the one tested; I can't tell which from the packet.

Requested: regenerate editor/BUILD.json from the shipped sources (node editor/build.mjs), and confirm that editor-provenance.test.mjs passes 6/6 under MOSAIC_REQUIRE_EDITOR_BUILD=1 on the tree the patch is cut from. Please include that log in the packet.

A correction to my own earlier note to Sage: I said no test compares the shipped bundle with BUILD.json. That was wrong; :79/:89 does, and it is the test that catches this.

Checked and correct

Fallback, my r1 change. editor-browser.test.mjs:382 stubs roundTrip to disagree. It checks the Source page, the note, Save off, and the Visual button opening the editor with no new read once the file round-trips. It also checks window.MosaicEditor = undefined. My g7 (docShown never set, bus.js) is killed by tests 4, 5 and 6.
docShown versus docMode. The click handler returns only when the mode equals what is on screen, so Visual from a fallback acts and the user's preference survives the next fallback, as the test asserts.
One-line sinks.
- oneLine (model.mjs:168) clamps the link title, image alt and image title.
- fenceInfo (:169) cuts the info at the first line break and drops an info holding a backtick.
- The four paste tests wait for the pasted node to hold the hostile value before they save. g1 to g4 are each killed by their own test (7 to 10).
Tilde fence with a backtick info. :150 sets refused, and the block stays as source. g6 is killed by editor-browser.test.mjs test 11.

Non-blocking notes

g5 survived. It removes the backtick drop in fenceInfo (:171, return info;). No test saves a block whose info holds a backtick on the save path; the load path refuses that case first (g6). A paste of <pre data-params="ab">` followed by a save would cover it.
Packet wording. BUILD.md :388, :390 and :419 say the tilde backtick case "opens as Source". The code and test 11 show a raw source block (pre.md-raw) inside the visual editor, not the Source view. Dewey's ruling on (ii) should be read with that in mind.
docs-browser.test.mjs:67. The second wait dropped the explicit #doc-commit existence check that :65 keeps, so undefined !== "" passes before the element exists. The #doc-text check in the same wait covers that in practice.
A title ending in a backslash (BUILD.md :394) is still unescaped for link and image titles. The save is corrupted but stays on one line, so no HTML block results. A follow-up is fine.

Ran by me

Slot from sage-0828. The busy check was CLEAR before each step. TMPDIR was in my own scratch, DOCKER_HOST was a dead socket, and MOSAIC_REQUIRE_EDITOR_BUILD=1 was set.

Rebuild, 18:28:45Z.
- npm ci --ignore-scripts in packages/webui, lock and binary shas as above.
- One build to a scratch --out, then one in place; the shipped files were restored afterwards and the manifest re-matched.
- Results as above.
Singles, 18:29:03Z to 18:32:09Z.
- Every packages/webui/tests/*.test.mjs file was run singly: 32 files, 199/201 by awk sum, 0 skipped.
- The two failures are the provenance tests above.
- I reran editor-provenance.test.mjs once with the spec reporter for the assertion text.
- All other files passed, including editor-browser 11/11, editor 5/5, docs-browser 1/1 and service 23/23.
Mutants, 18:32:32Z to 18:34:30Z.
- Both controls (editor-browser.test.mjs, editor.test.mjs) passed first.
- Each model.mjs mutant was rebuilt into the bundle before its run, and the files were restored after each one.
- 7 run, 6 killed (g1, g2, g3, g4, g6, g7), 1 survived (g5).

Not mine

Researcher's r2 runs in BUILD.md: two byte-identical builds of 32fcaa18...; singles 62/62 and webui 201/201; 3 mutants killed. I did not rerun them, and the provenance part of those counts doesn't match what I see on this candidate.

5.9 KiB Raw Permalink Blame History