feat: add durable execution work orders#8403
Conversation
|
Quick CI note on this draft follow-up:
So the current red CI does not appear to be introduced by this H007 follow-up itself. |
447625d to
85857d5
Compare
Refresh — April 14 2026Branch restructured and rebased onto current What changed in this refreshThe original branch had two commits stacked: an older version of
Test results
Commit logDependency noteThis PR builds on top of PR #9209 (execution receipts). The branch currently includes the #9209 commits inline; if #9209 merges first, this branch can be trimmed to just the CI notes
|
Stacking noteThis PR depends on #9209 (execution receipts). The current branch includes the #9209 commits inline so it is reviewable and testable in isolation, but the intended merge order is:
If you prefer to review both together before any merge, no action is needed — the branch is self-contained as-is. |
85857d5 to
7d9f653
Compare
|
Refresh concluído em branch limpa a partir de origin/main. Cherry-pick aplicado e push forçado para h007-durable-work-orders. Validação: py_compile nos .py alterados e pytest -o addopts='' -q nos testes afetados (13 passed). |
|
There is substantial real work here and the PR body includes strong local evidence:
So this is not a “throwaway” PR. However, it is still not merge-ready as packaged today. Main reasons:
Recommendation: KEEP OPEN, but treat it as blocked on packaging + dependency/acceptance decisions. Suggested next step:
So: likely still relevant, but currently too large and too dependent on unresolved base decisions to push for merge as-is. Batch 4 — technical follow-up required |
Fechando para Reavaliação ArquiteturalComo parte de uma auditoria completa do backlog (23 PRs abertas auditadas), esta PR foi identificada como dependente de clareza arquitetural upstream. Contexto: Motivo do fechamento:
O que acontece agora:
Agradecimento: Fechado como parte do audit de backlog 2026-04-19: 21 PRs refreshadas, 4 fechadas (2 absorvidas, 2 reavaliadas) |
Summary
queue/run_due,retry,resume,cancel, andreclaim_staleoperationsworkordersCLI/slash surfacesH007 line overview
This is part 2 of the same H007 line as #8402.
Intended review shape:
The architectural point is narrowness, not breadth:
delegate_taskdirect pathMotivation
Once the narrow direct lane is auditable, the next H007 step is to make it durable.
This PR adds a small control-plane slice around that lane:
It deliberately avoids inventing a second execution path.
What this PR changes
tools/execution_work_orders.pyqueued,running,retry_scheduled,completed,failed,cancelledtools/execution_work_orders_tool.pylist,query,enqueue,run_due,reclaim_stale,retry,resume,cancel,runner_status,install_runner,remove_runnerhermes workorders .../workorders ...direct_terminal_work_orderValidation
Focused local validation on this branch:
Result:
434 passed, 1 warningExternal real-run evidence for this queue layer on the supported slice:
4.37sduration /4.56swall0.20sduration /0.38swall0.22sduration /0.44swall94.97%faster duration90.35%faster wall30/30valid runs30/30exact-match runsReal scheduler/resumability validation also proved:
2run_dueNotes for reviewers
This PR is intentionally narrow and depends conceptually on #8402.
It does not claim:
The claim is narrower: