Follow-up: document concurrency/cancel-in-progress & fan-out semantics under synchronous dispatch (post ADR 0041)

[ifireball]

ADR 0041 is accepted; this issue captures follow-up doc/ADR work raised during review.

Problem

Moving from workflow_dispatch fan-out to workflow_call changes runtime semantics:

• Concurrency and cancel-in-progress behavior (shim/job held for full agent duration; potential queuing vs cancel).
• Fan-out/composition: stages become jobs in a single run; multiple independent workflow files per stage are no longer separate top-level runs.

What to do

• Describe the concurrency model we want (per-stage? per-run?) and where concurrency groups/cancel-in-progress should live after the change.
• Decide whether cross-stage cancellation is needed/possible (e.g. triage comment cancels review run).
• Document how multi-agent composition inside a stage should work given workflow_call constraints.

References

• PR docs: accept ADR 0041 and update living docs #1578 review thread about concurrency + fan-out.

[fullsend-ai-triage]

Triage: Blocked

This issue tracks follow-up documentation work (concurrency model, cancel-in-progress, fan-out semantics) that depends on the synchronous workflow_call dispatch model from ADR 0041.

However, PR #1578 — which accepts ADR 0041 and updates living docs — is still open. The ADR remains in "Proposed" status. Until that PR merges and the dispatch model is officially accepted, the concurrency and fan-out semantics described here don't have a stable foundation to build on.

Once PR #1578 lands, this issue will be unblocked and ready for a contributor to draft the follow-up ADR(s) or problem doc updates covering:

• Per-stage vs per-run concurrency group placement
• Cross-stage cancellation semantics
• Multi-agent composition under workflow_call constraints

---

Labels: Documentation follow-up about dispatch concurrency semantics, touching the dispatch area at low priority since it's blocked.

[ralphbean]

Concrete example of this just showed up on #1746 — wrote it up on #1611 (comment): #1611 (comment)