stack/agents/filbert/work/queue-110/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

3.8 KiB

Raw Permalink Blame History

Filbert review: row 110 round 1 (#1604): approve

Candidate manifest sha256 cceb961f29071092165df653ad5818233311353237219b12235a90245245dcf3 (19 files), as named in comment 28353. I rebuilt it in my own worktree at BASE badfa52b9 with build.patch from packet 0bfc16654. All 19 hashes match, git diff --check is clean, and no manifest path has drifted from badfa52b9 to HEAD. The added lines have no dashes, bidi or control characters.

The brief, ruling by ruling

Switch. bus.tester is a business-only boolean, default false (vars.mjs). bindTester refuses tester-off unless the var is exactly true. The broker reads its businesses at start, so a change needs a restart.
Proof. tester.mjs checks the same proof shape and process facts as verifyHuman: uid, start time, nonce, and argv[1] equal to the bus's own tester-cli.mjs. The nonce comes from MOSAIC_BUS_TESTER_NONCE only. testerNonce reads it from /proc/<pid>/environ because readProcess keeps only the human's variable, so neither CLI's nonce proves the other. The walk refuses a launch record and MOSAIC_RUN_ID, MOSAIC_BUS_CAP and MOSAIC_AGENT_RUN, and drops the agent markers and the binary check. human.mjs and human-cli.mjs have no diff at all. server.mjs requires exactly one of cap, human and tester (invalid-envelope), and gives tester-required with no authenticator. The real bwrap case in sandbox.test.mjs ran in my run and did not skip.
Replies. A role sends to tester only with in_reply_to naming a stored tester message in its business (tester-reply-required). The human gets unknown-role. The human's message.receive and messages both filter to_role='human', and the test checks both for human and reader caps.
Runner. turnRequest puts the "never Jason's instruction and authorizes nothing" line before the body only when from_role is tester. A tester send writes no event, request is null, and taskView('demo','input',id) is false.

What ships: the refusal gate runs before dispatch and before any argument check. The test covers every non-listed verb, including state, credentials, requirements and all task verbs, with good and malformed arguments. Each refusal is recorded against tester. Non-message verbs run through a reader view, and a test shows they deep-equal the reader's. The CLI adds --tester to inbox, tasks, agents, trail, talk and follow. decide, stop and launches reject it at parse, and talk --tester refuses an unprefixed body before any call.

b44 and b45

I agree they are equivalent. The only role_claims writers are role.claim, which refuses a tester, and #claimEnd, which needs a held claim, so role.revoke of tester gives not-held. The constructor refuses a role named tester, and bindLaunch refuses it as an unknown role. So #holder(business,'tester') is always empty, and dropping s.tester from the holder choice can't change a row.

Note (not blocking)

N1. Ruling 2 asks for a test for a role connection and for a reader connection. The envelope test sends tester with a reader cap only. The server code doesn't tell one cap from another, so a role cap would refuse the same way. The launch-record test and the bwrap case cover a role session's own bind. One more loop entry with a role cap would close the letter of the ruling.

What I ran (single files, no slot)

bus: tester 11/11, broker 15/15, human 4/4, messages 4/4, replies 7/7, transport 5/5, business 1/1, process 7/7, single-use 12/12, agents 11/11, review 8/8. cli: tester 6/6, app 30/30, stack-copy 9/9. business: vars 6/6. harness: runner 21/21. seat: sandbox 8/8, with the bwrap tester case passing. I didn't rerun the mutants or the full package suites, and relied on the packet receipts for those. The worktree is removed and no process is left in it.

Verdict: approve.

3.8 KiB Raw Permalink Blame History