fix: add custody-aware stall recovery - #159

Merged

dnth merged 21 commits into

mainfrom

fm/fm-stall-recovery-ship

Sep 24, 2026

Commits on Sep 24, 2026

feat: custody-checked bounded auto-recovery for stalled workers
dnth
committed
f40c1d6
fix: restore hermes busy adapter, clean shellcheck, fix bun skip
dnth
committed
149cf4f
no-mistakes(document): Document stall-recovery artifacts and OMP recovery behavior
dnth
committed
ba0772e
fix: close stall-recovery custody gaps from verification review

firstmate
authored and
dnth
committed
58d478b
fix: restore worker instructions on in-lock relaunch cancellation

firstmate
authored and
dnth
committed
3f6567b
no-mistakes(review): Fixed invalid task-ID journal path traversal
dnth
committed
1fdcf30
no-mistakes(review): Made recovery attempt bounds fail closed and atomic
dnth
committed
52a1d86
no-mistakes(review): Propagated overrides and failed closed on restore errors
dnth
committed
f2e1118
no-mistakes(document): Updated inbox recovery documentation and comments
dnth
committed
6d4eacc
no-mistakes(lint): Suppress intentional background-lock subshell SC2031 warnings
dnth
committed
4ed041a
fix: defer stall relaunch when worker goes busy; reset attempt bound per record
dnth
committed
998e56a
no-mistakes(review): Reject recovery markers containing trailing extra lines
dnth
committed
fbbf497
no-mistakes(document): Document supervised stall-recovery relaunch behavior
dnth
committed
d5c1042
fix: fail closed on unproven custody and non-canonical attempt markers
dnth
committed
f0b796f
feat: reshape stall recovery - live workers escalate untouched, missing endpoints relaunch
dnth
committed
c69e698
docs: scripts.md stall-recovery description matches the captain-ruled shape
dnth
committed
3ae520e
no-mistakes(review): Closed relaunch custody race and inbox read failures
dnth
committed
3166564
no-mistakes(review): Propagated unreadable inboxes to stale-wake escalation
dnth
committed
0738349
no-mistakes(review): Permit dead endpoints and recheck before relaunch
dnth
committed
01e0dfa
no-mistakes(document): Document absent-endpoint stall recovery behavior
dnth
committed
fd6134a
no-mistakes(ci): Fixed bin/fm-control.sh relaunch state handling: state is now captured for all relaunches and the custody guard runs after the stopping journal immediately before lifecycle control…
dnth
committed
55ac2c9

Navigation Menu

fix: add custody-aware stall recovery - #159

fix: add custody-aware stall recovery#159
dnth merged 21 commits into
mainfrom
fm/fm-stall-recovery-ship

Commits on Sep 24, 2026

feat: custody-checked bounded auto-recovery for stalled workers

fix: restore hermes busy adapter, clean shellcheck, fix bun skip

no-mistakes(document): Document stall-recovery artifacts and OMP recovery behavior

fix: close stall-recovery custody gaps from verification review

fix: restore worker instructions on in-lock relaunch cancellation

no-mistakes(review): Fixed invalid task-ID journal path traversal

no-mistakes(review): Made recovery attempt bounds fail closed and atomic

no-mistakes(review): Propagated overrides and failed closed on restore errors

no-mistakes(document): Updated inbox recovery documentation and comments

no-mistakes(lint): Suppress intentional background-lock subshell SC2031 warnings

fix: defer stall relaunch when worker goes busy; reset attempt bound per record

no-mistakes(review): Reject recovery markers containing trailing extra lines

no-mistakes(document): Document supervised stall-recovery relaunch behavior

fix: fail closed on unproven custody and non-canonical attempt markers

feat: reshape stall recovery - live workers escalate untouched, missing endpoints relaunch

docs: scripts.md stall-recovery description matches the captain-ruled shape

no-mistakes(review): Closed relaunch custody race and inbox read failures

no-mistakes(review): Propagated unreadable inboxes to stale-wake escalation

no-mistakes(review): Permit dead endpoints and recheck before relaunch

no-mistakes(document): Document absent-endpoint stall recovery behavior

no-mistakes(ci): Fixed bin/fm-control.sh relaunch state handling: state is now captured for all relaunches and the custody guard runs after the stopping journal immediately before lifecycle control…