[DRAFT] fix(tools): support delegation reasoning overrides for subagents - #5337
Conversation
|
Follow-up from the same GitHub Copilot / ACP audit. This one is more logically separable than the ACP session/run-agent fixes, but it was validated locally on top of the same rebased stack. Related PRs:
|
a08f33c to
ffb1ff4
Compare
ffb1ff4 to
6576751
Compare
|
Closing as absorbed by current main. I attempted a clean refresh of this branch against current , and the rebase conflicted specifically because the main branch already contains the delegation reasoning-effort behavior and its dedicated test coverage. I then validated the current upstream surface directly in a clean tmux-based run: |
|
Correcting the closure note with the exact evidence: I attempted a clean refresh of this branch against current I then validated the current upstream surface directly in a clean tmux-based run:
So this PR no longer represents a needed delta on top of current main. |
This PR is logically independent from the ACP adapter fixes.
It came out of the same GitHub Copilot / delegation audit, and I’m opening it as a draft early to reduce the risk of upstream drift.
Summary
This PR improves delegation behavior so subagents can use explicit reasoning settings instead of blindly inheriting the parent agent’s reasoning configuration.
What changed
delegation.reasoning_effortacp_command/acp_args) while wiring reasoning overrides on topWhy
Delegation should be predictable and configurable, especially when the parent agent is using a different reasoning profile than the intended worker/subagent lane.
This makes subagent routing more explicit and less surprising.
Validation
Automated
python3 -m py_compile tools/delegate_tool.py hermes_cli/config.py tests/tools/test_delegate.pyuv run --extra dev python -m pytest -o addopts='' tests/tools/test_delegate.py -q -k 'DelegationReasoningEffort'2 passed, 51 deselectedNotes
tools/delegate_tool.pyhas evolved since the original branch, specifically around ACP delegation transport overrides.