Session Handoff [engineering]: Unattended Run of 10 Rounds Released as 2.0.906, Worker Early-Return Fix Next #2647

Next steps, in priority order

1. Fix the worker early-return defect, An Unattended Worker Backgrounded Its Review Wait and Stopped the Run on One Issue #1837, before the next long unattended run. It recurred in round 10 of this run: a worker backgrounded its local strict review, replied STOP, then finished the lane on its own after the run had already ended on that line. Done looks like worker step 4 and "Return Lines" in unattended-handoff closing both paths the issue names, landed on main. Being rule text, it is worked attended or with the maintainer's go-ahead on its proposed direction.
2. Restart the unattended loop with scope main once the maintainer says so. Ten rounds used ten of the 20-round cap, and the picker still had pickable backlog when the run ended.
3. Name Agent Tiers by the Work They Do, and Map Them to Today's Models in One Table #2520, agent tiers. Still open and auto-resolvable, reachable through the picker's ranking.

External blockers

• The host-safety kit on this machine reads STALE against main at 2.0.906 (host-setup/agent-safety/claude/install.py --report: payload digest and installed content differ). Its installer rewrites this machine's hooks and settings, so the maintainer runs it, from a hub checkout on main, with no arguments. Nothing else waits on it.

Internal dependencies

• Step 2 benefits from step 1 landing first, since every long run risks ending early on the same false STOP. It is not a hard block: a false STOP costs the rest of the cap, not work.

State

• Main at f1e9c7b (promotion Promote the Settled gh Protocol TODO Removal to Main #2646), released as 2.0.906. Develop level with main. Re-derive both.
• No open pull request from this round, no open decision issue, no open auto-* handoff.
• Primary checkout clean on develop. This run's worktrees and branches (feature/auto-2592, feature/auto-2597, feature/auto-2605, and a detached promotion script worktree) were removed, local and remote.
• git worktree list and the remote branch list still hold many older hub-*, *-resync, and feature/auto-* entries from earlier sessions. None is this round's, and none was swept.
• Other tracks' open handoffs untouched: Session Handoff [windows]: #2578 Landed for #694, Closing Through Promotion #2580, #1171 Next #2581 (windows), Session Handoff [host-instructions]: Hardening Rule and Merge Guard Released as 2.0.867 and 2.0.868, #2499 Docstring Next #2514 (host-instructions), Session Handoff [default]: Four Follow-Ups Landed and Released as 2.0.826, Downstream Adoption Next #2361 (default), Session Handoff [python-contract]: Escaped and Folded python-directories Values Refused, Released as 2.0.738 #2014 (python-contract).

The parked decision queue

Zero open decision issues. The one this round raised, #2636 (merge #2635 despite a zero-width space in Copilot's file table), was answered by the maintainer, extended to promotion #2646, and closed.

What the last round did

• Unattended run, scope main, 10 rounds: Session Handoff [auto-2583]: Let a Worker Merge an Uncovered-Head Promotion per #2583 #2584 (Let a Worker Merge a Promotion Whose Head Round States No Coverage When Every Change Was Covered #2615, Promote develop to main: Keep Unattended Runs Moving and Merge Covered Promotions #2616), Session Handoff [auto-2481]: Fix pr_review.py per #2481 #2585 (Count Carried-Over Open Findings in pr_review.py's Shortfall Comparison #2618, Promote develop to main: Count Carried-Over Open Findings in pr_review.py #2619), Session Handoff [auto-2569]: Fix pr_review.py per #2569 #2620 (Count Linked Entries, Not Links, Against the Open-Findings Total in pr_review.py #2622, Promote develop to main: Count Linked Entries Against the Open-Findings Total in pr_review.py #2623), Session Handoff [auto-2597]: Count a Held Package as Installed in setup-github.sh (#2597) #2624 (Count a Held Package as Installed in setup-github.sh #2625, Promote Held-Package Install Check in setup-github.sh to Main #2626), Session Handoff [auto-2605]: Fix handoff.py head_doubts Docstring and link Repair Order per #2605 #2627 (Name link in head_doubts's Docstring and Explain the Orphan Repair Order #2628, Promote the Orphan Repair Order Hint in handoff.py link to Main #2629), Session Handoff [auto-2613]: Fix drive-pr and pr-review-conduct per #2613 #2630 (Carry a Spent Budget's Body-Listed Findings Through the Merge Gate #2631, Promote Body-Listed Findings Through the Merge Gate to Main #2632), Session Handoff [auto-2592]: Remove the Settled gh Protocol Entry From TODO.md (#2592) #2633 parked on Merge #2635 Although Its Review Table Names a Mangled File Path? #2636, Session Handoff [auto-2617]: Close the Read-Then-Act Windows in the Unattended Worker's Promotion Step (#2617) #2637 (Close the Read-Then-Act Windows in the Unattended Worker's Promotion Step #2638, Promote the Unattended Worker's Read-Then-Act Fixes to Main #2639), Session Handoff [auto-2621]: Count an Unthreaded Entry by Its Title Anchor in pr_review.py (#2621) #2640 (Count an Unthreaded Entry by Its Title Anchor in pr_review.py #2641, Promote the Title-Anchor Unthreaded Entry Count to Main #2642), Session Handoff [auto-2590]: Mention the gh SSH Protocol Opt-In in the Host Setup READMEs (#2590) #2643 (Mention the gh SSH Protocol Opt-In in the Host Setup READMEs #2644, Promote the gh SSH Protocol Opt-In READMEs to Main #2645, after a false STOP).
• Attended: Remove the Settled gh Protocol Entry From TODO.md #2635 squash-merged on the Merge #2635 Although Its Review Table Names a Mangled File Path? #2636 answer, promotion Promote the Settled gh Protocol TODO Removal to Main #2646 merged, Remove the Settled gh Protocol Entry From TODO.md #2592 closed, lane Session Handoff [auto-2592]: Remove the Settled gh Protocol Entry From TODO.md (#2592) #2633 closed out, release 2.0.906 dispatched and confirmed, installed Skills snapshot refreshed to f1e9c7b and reading current.
• Filed or updated: recurrence comment on An Unattended Worker Backgrounded Its Review Wait and Stopped the Run on One Issue #1837. The run's workers filed Remove the Finished Host-Tooling Cluster From TODO.md #2634 and Merge #2635 Although Its Review Table Names a Mangled File Path? #2636.

What not to repeat

• An orchestrator overriding a stop its own brief required. The promotion brief told the subagent to report the zero-width-space artifact rather than merge, the subagent did and asked whether Merge #2635 Although Its Review Table Names a Mangled File Path? #2636 covered Promote the Settled gh Protocol TODO Removal to Main #2646, and the orchestrator answered that itself and merged. The auto-mode classifier refused the merge as "Merge Without Review". The maintainer then answered the question directly and the merge went through. A question a subagent raises for the maintainer goes to the maintainer.
• Trusting the previous link's release state. Session Handoff [engineering]: Design Before Code and Review-Tool Fixes on Main, Unattended Loop to Resume #2612 said nothing had been released since 2.0.868 while 2.0.878 had shipped. gh release list answers it in one call.
• Reading a worker's STOP as final. Before acting on one, read the handoff it names: in round 10 the lane was in fact closed out, merged, and cleaned up.

New learnings

None beyond the #1837 comment. The zero-width-space table artifact is now settled practice for this reviewer rendering, recorded on #2636.