jason.woltjeandClaude Opus 5.5
1e3c06c78b
Round 2 of #1519: R1 to R5 fixed and reproduced, 43/43 on Node 26 and Node 24, 22 of 23 mutants killed against a clean baseline. Records a reparented-CLI path past the human proof for Sage's ruling. Co-Authored-By: Claude Opus 5.5 <[email protected]>
1.5 KiB
Row 37, S2, round 2: approve (Darkwing)
Candidate candidate-manifest.sha256 61519059…, 24 files under
packages/bus/. Full record:
agents/darkwing/work/slice1-s2-review/review-r2.md.
The patch applies at fef4b362 and adds only packages/bus/. The
manifest checks 24/24. Tests pass 43/43 on Node 26 and on Node 24, and
Rocko's contract script passes against S1 round 2.
All five required items are fixed, and I reran my round 1 probes:
- R1: terminal 8357's real ancestry passes. Unreadable ancestors still get the command and launch checks.
- R2: a revoked run gets
run-revoked, including after a broker restart. - R3: cto and pm both route their own cross-role decisions to
human. - R4: a refused bind replies and the broker keeps serving.
- R5: 17 of Rocko's mutants, plus 5 of my 6, are killed against a clean baseline, and the no-op isn't.
For Sage: since R1, an agent run here passes the human proof with
setsid -f env -i node human-cli.mjs, because systemd --user's environ
is unreadable. I reproduced it from my own session. A CLI reparented to
pid 1 passed in round 1 too. That's inside the README's cooperative
limit, but REQ-DEC-3 doesn't say so. I recommend a lead decision that
slice 1's human proof is cooperative, with S6 isolating managed agents.
My round 1 claim that the fix wouldn't weaken anything was wrong.
Non-blocking: no test checks a within-role route_to, which is my one
surviving mutant. Two R2 contract lines come from a script that isn't in
the packet. A refused bindLaunch uses up its run name.