Everything I could decide myself is decided. What's left needs
you. Each one has a recommendation from Fable — if you agree, pick it and
press Done. "I'm not sure" is a real answer and becomes work for me.
You don't need to open the session — pressing Done sends your answer back and it picks up.
1The fleet-health read is built and committed (4 commits, local only). Its one migrated caller fixes a live defect I measured: 39 task files right now tell an operator to 'respawn' a Worker that is running fine — a subagent has no roster row, so the absence it reads as death is the designed state, and acting on it would start a second Worker beside a live one. The fix does nothing until ./deploy.sh runs, and that restarts BOTH live services (the relay serving /agents and the supervisor keeping ~43 sessions alive) from code that is unpushed, unreviewed and has never been through CI. How do you want it shipped?
NARROWING THE RISK, measured after raising this: the change touches the RELAY package only — 4 files of source plus tests and docs. The SUPERVISOR package (the one that keeps every session alive, holds the red-main claim and runs the compact sweeps) has NO new code in these commits at all. deploy.sh still restarts both, so a supervisor restart is still a restart — but it would come back on the same code it is running now. The only new code going live is in the relay, which serves the agents page and writes the task pages.
Verified by the operator, not taken on the worker's word: relay typechecks clean; the health, route-guard, read-only, loop-state, digest and task-state suites pass (115 tests); an independent positive control written from the source shapes caught all three contradictions on a lying session while staying silent on two healthy ones, and returned 'unknown' rather than 'ok' when nothing could be measured. The full suite and the review gate have NOT run — that needs the push.
In plain terms. I finished the health check you asked for — one place that answers "is this agent actually alive and getting somewhere", instead of me piecing it together from four places every time I look.
While testing it I found something already broken. Right now 39 of your task pages tell whoever reads them to restart an agent that is working perfectly fine. The page looks for the worker in a list it was never going to appear in, doesn't find it, and concludes it died. If someone follows that advice they start a second agent on top of a running one. My fix corrects it, and it also fixed my own task's page, which was telling you the same lie about the agent that did this work.
The catch: none of it takes effect until I restart the two services that run the whole fleet — the one behind the agents page, and the one that keeps every session alive. Nobody has reviewed this code yet. If I restart them and something is wrong, the thing that would normally recover it is the same thing I just restarted.
So: do you want the normal careful route (I push it, it gets reviewed, then we turn it on), or do you want it live tonight because 39 wrong pages is 39 too many?
Fable recommends: Normal path: I push the branch and open a PR, it gets reviewed and merged, then deploy The false text is a pre-existing condition, not tonight's regression, and it only causes harm if someone acts on it — so the hours a review costs are affordable. Deploying unreviewed code to the supervisor is the one machine that would otherwise revive things if it broke, and the repo added its review gate three days ago precisely because unreviewed work was reading green. Fable was asked, and escalated. Receipt f3a3fc4a4 — NEEDS HUMAN. Both passes agreed on 'Hold the deploy — report 4 commits ready and the measured defect, let the human authorise push/review/merge/deploy', and it is being escalated anyway: this question turns on money or a commitment to a person (matched 'committed'), which is not Fable's to settle however confident it sounds.
Paid for on 2026-08-07: the unit-607 renewal rate came back RESOLVED from both passes at high confidence and was wrong — Fable found the standup ruling about the move-in SPECIAL and missed the correction posted 2h21m later saying she had actually asked about the RATE.
Fable's reasoni…
635a32e6 is parked on this. 1 other session waiting on this same answer.
Asked by more than one session. Your answer above goes back to all of them.
10ea687d asked it this way: Push the vendor contact refresh branch so it can be reviewed and merged?