What Does This PR Do?

This PR makes delegate_task(profile=...) reliably launch a profile-backed Hermes ACP worker under the selected profile's runtime/home instead of silently falling back to the parent runtime.

It also tightens delegated-worker isolation:

• Delegated profile-backed ACP workers now skip context-file injection and memory injection.
• Child toolsets are derived from the parent's effective tool surface, but blocked composite toolsets are stripped so blocked tools cannot leak indirectly.
• ACP-backed child runtime metadata and usage handling are normalized so the parent/runtime surfaces stay truthful.

This is the right approach because it makes named-profile delegation explicit and reliable without changing the behavior of normal non-profile delegation.

Related Issue

No dedicated issue for this change.

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Security fix
• Documentation update
• Tests (adding or improving test coverage)
• Refactor (no behavior change)
• New skill (bundled or hub)

Changes Made

• tools/delegate_tool.py
  • Make delegate_task(profile=...) build a profile-backed Hermes ACP child reliably.
  • Derive child toolsets from the parent's effective toolsets.
  • Strip blocked composite toolsets, including messaging, so send_message cannot leak into children.
• run_agent.py
  • Preserve profile/transport/launch metadata on delegated children.
  • Keep child ACP/profile runtime state aligned with the selected worker profile.
• acp_adapter/session.py
  • Honor delegated worker env overrides and preserve worker isolation.
  • Force delegated profile-backed workers to skip context files and memory.
• agent/copilot_acp_client.py
  • Normalize ACP usage/runtime metadata surfaced back to Hermes.
• tests/tools/test_delegate.py
  • Add coverage for profile-backed delegate routing and blocked composite toolset stripping.
• tests/acp/test_session.py
  • Add delegated-worker env/toolset isolation coverage.
• tests/agent/test_copilot_acp_client.py
  • Add ACP usage/runtime metadata coverage.
• tests/run_agent/test_run_agent.py
  • Add end-to-end delegate/profile runtime coverage.
• Documentation updates:
  • website/docs/guides/delegation-patterns.md
  • website/docs/reference/tools-reference.md
  • website/docs/user-guide/features/delegation.md

How to Test

1. Run the targeted delegate/profile test slice:

   unset API_SERVER_KEY API_SERVER_HOST API_SERVER_PORT HERMES_HOME
   uv run --extra dev python -m pytest -o addopts='' \
     tests/tools/test_delegate.py \
     tests/acp/test_session.py \
     tests/agent/test_copilot_acp_client.py \
     tests/run_agent/test_run_agent.py -q

2. In a Hermes session with any existing named profile, invoke delegate_task(profile="<your-profile>", toolsets=["terminal","file"]).

3. Confirm the delegated worker:

   • Launches via hermes --profile <your-profile> acp.
   • Uses the selected profile's runtime/home.
   • Does not expose blocked child tools such as delegate_task, clarify, memory, execute_code, or send_message.

Checklist

Code

• I've read the Contributing Guide.
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.).
• I searched for existing PRs to make sure this isn't a duplicate.
• My PR contains only changes related to this fix/feature (no unrelated commits).
• I've run pytest tests/ -q and all tests pass. See full-suite note below.
• I've added tests for my changes (required for bug fixes, strongly encouraged for features).
• I've tested on my platform: Ubuntu 24.04.3 LTS.

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) - or N/A.
• I've updated cli-config.yaml.example if I added/changed config keys - or N/A.
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows - or N/A.
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide - or N/A.
• I've updated tool descriptions/schemas if I changed tool behavior - or N/A.

Screenshots / Logs

Targeted delegate/profile slice:

370 passed in 18.77s

Local disposable delegate-profile smoke also passed after rebasing onto current main.

Full-suite note:

• Current origin/main is baseline-red.
• I did run a branch-vs-main full-suite comparison after rebasing onto current main.
• Raw full-suite counts still differ, but spot-checking the branch-only failures individually on both the rebased branch and current main did not confirm a branch-only regression.