stack/agents/filbert/work/queue-108/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

4.7 KiB

Raw Permalink Blame History

Filbert review: row 108 round 1 (#1602): approve

Candidate manifest sha256 19ef7133db592787c24fa473357687a5741ccd7823627b970624c930a7611dbc (6 files), as named in request 28485. I rebuilt it in my own worktree at BASE 234121b30 with build.patch from packet a6e4fe747. All 6 hashes match, and git diff --check is clean. No manifest path has drifted from 234121b30 to HEAD. The patch's added lines have no dashes, bidi or control characters. All six paths are in the brief's file list.

The brief, item by item

Drop-in guard. The existing loop over systemd-analyze --user unit-paths now globs mosaic-bus@*.service.d, mosaic-.service.d and service.d in each directory, with the same refusals on an unreadable directory or file. Those are the only prefix drop-ins systemd applies to mosaic-bus@<b>.service: the name has one dash, so the only prefix is mosaic-. The new tests put WorkingDirectory and then ExecStart in each inherited directory and get exit 1 with the file named and the old unit bytes unchanged. An Environment=-only file is still allowed. A dangling directory symlink and a dangling .conf symlink both refuse before writing.
Header comment. The render and install lines now carry exactly the flags in USAGE: --copy ID [--copies DIR], --pm, --dir, --no-reload and --claude.
Dead show arms. The packet says it removed them. The install fakes in host.test.mjs and stack-copy.test.mjs now answer list-units and exit 99 on anything else, so any unexpected systemctl call by install now fails the test, which is stricter than before. The systemd 261 template-rejection arm stays in its test. If install ever ran show [email protected] again, that arm would fail it, and so would the calls log check. The status fake's show arm is reached by status.
Wrapper. The marker stays on line 2. Two comments follow it, naming the copy root and saying the Q81 lease probe answers at mosaic-old. Then come the exact lease line and the final exec. $root has already passed plain_path, so the comment can't break the file. A rerun stays idempotent: the marker is still on line 2, and the lease line still matches once under grep -Fxc. The test asserts lines 2 to 4 and the last line, runs the lease, and shows check passing on the new wrapper. The foreign-wrapper case now has matching effective roots and a matching exec, so the marker alone refuses it. That is the fixture fix the packet records after the first marker mutant survived.
Install-script audit. I checked it independently. No file in COPY_PATHS mentions @google/genai, esbuild or protobufjs. PI_BIN in pi-pin.mjs and pi's bin both select dist/bundle/cli.js, and the bundle has no reference to @google/genai or protobufjs. Its only esbuild text is undici's esbuildDetection variable, not a load. The unbundled pi-ai Google providers do import @google/genai, but nothing on the copy's paths runs them. The README paragraph says the warning is expected and why, and that a new runtime loader needs a fresh audit. There is no allowScripts change, so the row doesn't stop for Sage.

Mutants

The packet's seven mutants cover each new check: miss-prefix, miss-all-services, ignore-unreadable-directory, ignore-unreadable-file, omit-copy-comment, omit-lease-comment and accept-foreign-check-wrapper. I read mutants.py and the final out/mutants/*.txt. Each mutant is a unique string replacement restored in finally, and each fails the case written for it. Reading the tests, I'd expect each one to be killed for the stated reason. For example, with the file-read refusal removed, the dangling .conf reads as empty, install exits 0, and the test's status 1 fails.

Notes, no change needed

The new template-261 assertion runs the fake systemctl directly. That tests the fixture, not bus-service.sh. It's harmless, and the arm still guards a regression as described in item 3.
The inherited-directory tests use the first lookup directory only. The glob list is inside the per-directory loop, and the existing two-directory test covers that loop.
The render line in the header comment is padded past the column the other lines use.

What I ran

I ran the rebuild and the checks above. In my own worktree, with DOCKER_HOST pointing at a dead socket, I ran single test files without the slot: node --test packages/cli/tests/host.test.mjs passed 41 of 41, and packages/cli/tests/stack-copy.test.mjs passed 12 of 12. I did not rerun the mutants, the full CLI suite or the shell suites. Those counts are Rocko's receipts (CLI 187, 7 of 7 killed). The Docker-dependent steps Rocko skipped run in Sage's landing gate.

Verdict: approve.

4.7 KiB Raw Permalink Blame History