fix(auxiliary): rebuild closed cached sync clients

[alex107ivanov]

What does this PR do?

Evicts cached auxiliary clients whose underlying sync HTTP transport is already closed before reusing them. Without this guard, a closed cached sync client can be returned from _get_cached_client, causing follow-up auxiliary calls to keep failing until process restart instead of rebuilding the provider client.

Related Issue

Related to auxiliary client cache/transport lifecycle fixes; no dedicated issue found for this exact closed-sync-cache-hit path.

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

• Added _client_is_closed() in agent/auxiliary_client.py to detect cached clients or wrapped clients whose transport exposes is_closed is True.
• Updated _get_cached_client() to evict closed cached clients and rebuild them instead of returning them.
• Added regression coverage in tests/agent/test_auxiliary_client.py proving a closed cached sync client is replaced with a fresh client.

How to Test

1. PYTHONPATH=$PWD /workspace/repos/hermes-agent/.venv/bin/python -m pytest tests/agent/test_auxiliary_client.py::TestAuxiliaryClientPoisonedCacheEviction -q -o 'addopts='
2. PYTHONPATH=$PWD /workspace/repos/hermes-agent/.venv/bin/python -m pytest tests/agent/test_auxiliary_client.py -q -o 'addopts='
3. ln -sfn /workspace/repos/hermes-agent/.venv .venv && scripts/run_tests.sh tests/agent/test_auxiliary_client.py && rm -f .venv
4. PYTHONPATH=$PWD /workspace/repos/hermes-agent/.venv/bin/python -m py_compile agent/auxiliary_client.py tests/agent/test_auxiliary_client.py
5. Full scripts/run_tests.sh was attempted in this environment but timed out after 10 minutes, with failures observed in unrelated CLI/gateway systemd tests before timeout.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(auxiliary): ...)
• I searched for existing PRs to make sure this isn't a duplicate (related poisoned-cache/transport lifecycle fixes reviewed; no duplicate closed-sync-cache-hit fix found)
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run pytest tests/ -q and all tests pass (full suite attempted; blocked/timed out as noted above)
• I've added tests for my changes (closed cached sync client rebuild regression coverage)
• I've tested on my platform: Linux 5.15 container

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — N/A, internal cache lifecycle fix with no user-facing behavior/config change
• I've updated cli-config.yaml.example if I added/changed config keys — N/A, no config keys changed
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — N/A, no contributor workflow changes
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — platform-agnostic Python cache handling; no OS-specific paths/subprocesses
• I've updated tool descriptions/schemas if I changed tool behavior — N/A, no tool schemas changed

Screenshots / Logs

Targeted regression failed before the fix on upstream main:

FAILED tests/agent/test_auxiliary_client.py::TestAuxiliaryClientPoisonedCacheEviction::test_get_cached_client_rebuilds_closed_sync_client
AssertionError: assert closed_client is fresh_client

After the fix:

8 passed, 1 warning in 6.90s
208 passed, 1 warning in 20.96s
scripts/run_tests.sh tests/agent/test_auxiliary_client.py: 208 tests passed

[alex107ivanov]

Closing this PR as superseded by #39790. The replacement PR has been rebased onto current upstream main, force-pushed with lease, and has fresh focused validation evidence posted there.