fix: keep API parent runs stoppable and nonterminal during child progress #387

Merged

karsten merged 1 commit from fix/api-progress-is-not-completion into main 2026-11-13 03:57:08 +00:00

Conversation 1 Commits 1 Files changed 4 +42 -1

karsten commented 2026-11-13 03:53:18 +00:00

Owner

Copy link

Clean native parallel test exposed a second reporting defect: child started status became the API parent final output and completed it before the real parent reply; stop then returned409 although work remained. Treat explicit presentation progress as bounded SSE status.updated, not finish; all Dispatcher non-result notices addressed to API are progress, with Matrix attention semantics unchanged. Real parent answer/outcome and blocked maintenance retain terminal ownership. Late progress cannot mutate/resurrect terminal output.

Build/full 2106 passed,1 skip,0 failures; independent supplied-diff review APPROVE. Regression tests actual stop endpoint remains usable after child status and real final answer remains stored. This prevents a status line being counted as successful work, complementing native ready/continuation fixes. No permission/budget/replay changes, no installed-host suites.

Clean native parallel test exposed a second reporting defect: child started status became the API parent final output and completed it before the real parent reply; stop then returned409 although work remained. Treat explicit presentation progress as bounded SSE status.updated, not finish; all Dispatcher non-result notices addressed to API are progress, with Matrix attention semantics unchanged. Real parent answer/outcome and blocked maintenance retain terminal ownership. Late progress cannot mutate/resurrect terminal output. Build/full **2106 passed,1 skip,0 failures**; independent supplied-diff review APPROVE. Regression tests actual stop endpoint remains usable after child status and real final answer remains stored. This prevents a status line being counted as successful work, complementing native ready/continuation fixes. No permission/budget/replay changes, no installed-host suites.

karsten added 1 commit 2026-11-13 03:53:18 +00:00

fix: dispatcher progress must not complete or disable stop on API parent runs 5786c067b6

karsten merged commit ef740f5e34 into main 2026-11-13 03:57:08 +00:00

karsten referenced this pull request from a commit 2026-11-13 03:57:10 +00:00

Merge pull request 'fix: keep API parent runs stoppable and nonterminal during child progress' (#387) from fix/api-progress-is-not-completion into main

karsten referenced this pull request 2026-11-13 04:29:38 +00:00

fix: missing optional Web modules must not abort agent startup #381

karsten referenced this pull request 2026-11-13 04:29:38 +00:00

fix: re-assess current retained non-decision blocks without replay grants #386

karsten commented 2026-11-13 04:29:38 +00:00

Author

Owner

Copy link

Fleet activation is now actually verified for all13 active agents at ef740f5e34 (published container0.1.273, sha256:a728a04b09e84d92b02573465fce507c10314ac4c0bb99f5cc9ad43f71d22f28): five native +Grace/Minerva/Che/Kakashi/Minato through authenticated running APIs, Moist/Odo/Randy through actual /proc Node entrypoint resolving to immutable package CLI and same-root build metadata. All8 Kubernetes Deployments ready1/1; intentionally inactive broker remains0. Normal GitOps owner PRs; no task/state/config ledger surgery.

Behavioral evidence remains separate from version readiness: retained Shikamaru Matrix task automatically advanced7→11 with SDK tool receipts, uncancelled native worker overlap27.7196s, explicit stop still stopped/depth0/no pending after final restart. On final #387 revision, new live API probe run_b4c89ab6b2d243dc8ce17784bb5818f6 stores real parent final output including both accepted worker UUIDs, not started-status text; both read-only workers completed.

Persistent evidence index: /home/karsten/.local/share/hayate-lifecycle-work/FLEET-273-VERIFICATION.md; snapshot fleet-recovery-snapshot-1794543898.json. This proves fleet code activation and stated live cases, NOT that every retained task on every agent is presently runnable or that semantic/manual/legacy holds may be lifted.

Fleet activation is now actually verified for all13 active agents at ef740f5e345c (published container0.1.273, sha256:a728a04b09e84d92b02573465fce507c10314ac4c0bb99f5cc9ad43f71d22f28): five native +Grace/Minerva/Che/Kakashi/Minato through authenticated running APIs, Moist/Odo/Randy through actual /proc Node entrypoint resolving to immutable package CLI and same-root build metadata. All8 Kubernetes Deployments ready1/1; intentionally inactive broker remains0. Normal GitOps owner PRs; no task/state/config ledger surgery. Behavioral evidence remains separate from version readiness: retained Shikamaru Matrix task automatically advanced7→11 with SDK tool receipts, uncancelled native worker overlap27.7196s, explicit stop still stopped/depth0/no pending after final restart. On final #387 revision, new live API probe run_b4c89ab6b2d243dc8ce17784bb5818f6 stores real parent final output including both accepted worker UUIDs, not started-status text; both read-only workers completed. Persistent evidence index: /home/karsten/.local/share/hayate-lifecycle-work/FLEET-273-VERIFICATION.md; snapshot fleet-recovery-snapshot-1794543898.json. This proves fleet code activation and stated live cases, NOT that every retained task on every agent is presently runnable or that semantic/manual/legacy holds may be lifted.

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

harness/core!387

No description provided.