This issue is staged for the repository's GitHub tracker cutover. Existing task instructions remain authoritative until the verified handoff.
What
Observe a real human approving a requested session for each supported host, and that session acknowledging one real, harmless, throwaway first task, since the merged implementation explicitly states this live acceptance is still outstanding.
Why
The change that shipped this (letting the conductor request, and a human approve, a new session) passed two rounds of cross-model implementation and review that found and fixed five real defects, but its own pull request is explicit that this is not the same as live acceptance: "Live acceptance is still outstanding: explicit human approval for each supported host, observed captured project, model and isolation behavior, denial and admission bounds, and a harmless first-task acknowledgement. Automated lifecycle checks do not establish those outcomes. The task remains in Review." That closing line is the shipping change's own author concluding review status should hold, which is why the underlying task record correctly still reads review rather than done.
Scope
In scope: for each supported host kind, requesting a session against a disposable test project, approving it as a human, confirming the captured project, model and isolation behavior are correct, and confirming the newly launched session acknowledges one real, harmless, throwaway dispatched task (not real production work). In scope: confirming a denial starts nothing, and that the request and admission ceilings behave as designed. Out of scope: any code change; this is a verification pass over already-shipped code. Out of scope: dispatching any task with real consequence as the "harmless first-task acknowledgement," and performing this against a live, in-use workspace's real dispatch budget.
Done when
For each supported host, a human has approved one requested session against a disposable test project, observed correct captured project, model and isolation behavior, and observed that session acknowledge one real, harmless, throwaway dispatched task; a denial has been observed to start nothing; the result is recorded against this task and its status is updated to reflect the outcome.
Validation
Direct manual observation of real approvals, real launches, and one harmless first-task acknowledgement, once per supported host, against disposable test fixtures; this is exactly the evidence category the shipping PR's own text says is still missing.
Dependencies and decisions
None; the shipped code is not in question, only whether the human-approval-to-first-task path has actually been observed running for each host.
Links
Pull request #49, which shipped the underlying change and discloses this limitation directly, including the "task remains in Review" conclusion.
This issue is staged for the repository's GitHub tracker cutover. Existing task instructions remain authoritative until the verified handoff.
What
Observe a real human approving a requested session for each supported host, and that session acknowledging one real, harmless, throwaway first task, since the merged implementation explicitly states this live acceptance is still outstanding.
Why
The change that shipped this (letting the conductor request, and a human approve, a new session) passed two rounds of cross-model implementation and review that found and fixed five real defects, but its own pull request is explicit that this is not the same as live acceptance: "Live acceptance is still outstanding: explicit human approval for each supported host, observed captured project, model and isolation behavior, denial and admission bounds, and a harmless first-task acknowledgement. Automated lifecycle checks do not establish those outcomes. The task remains in Review." That closing line is the shipping change's own author concluding review status should hold, which is why the underlying task record correctly still reads review rather than done.
Scope
In scope: for each supported host kind, requesting a session against a disposable test project, approving it as a human, confirming the captured project, model and isolation behavior are correct, and confirming the newly launched session acknowledges one real, harmless, throwaway dispatched task (not real production work). In scope: confirming a denial starts nothing, and that the request and admission ceilings behave as designed. Out of scope: any code change; this is a verification pass over already-shipped code. Out of scope: dispatching any task with real consequence as the "harmless first-task acknowledgement," and performing this against a live, in-use workspace's real dispatch budget.
Done when
For each supported host, a human has approved one requested session against a disposable test project, observed correct captured project, model and isolation behavior, and observed that session acknowledge one real, harmless, throwaway dispatched task; a denial has been observed to start nothing; the result is recorded against this task and its status is updated to reflect the outcome.
Validation
Direct manual observation of real approvals, real launches, and one harmless first-task acknowledgement, once per supported host, against disposable test fixtures; this is exactly the evidence category the shipping PR's own text says is still missing.
Dependencies and decisions
None; the shipped code is not in question, only whether the human-approval-to-first-task path has actually been observed running for each host.
Links
Pull request #49, which shipped the underlying change and discloses this limitation directly, including the "task remains in Review" conclusion.