What does this PR do?

Unblocks shared PR checks that are currently red on main and inherited by otherwise mergeable PRs.

Latest base signal used for triage: main at fef1a41248a9a584f7b945d0a46d57de46d15358, failed Tests run 25612702683. I reproduced the current failed nodes locally on a clean worktree, then rebuilt this PR from current origin/main so it is one focused Stephen-owned unblocker lane.

This update also absorbs the useful parts of #22760 with commit attribution preserved via cherry-pick -x.

Related Issue

N/A. Base-CI unblocker for shared failures blocking multiple PRs.

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

• .github/workflows/lint.yml: grant issue-comment permission for same-repo lint summary updates and make the advisory PR comment step non-blocking, so fork token comment failures do not fail an otherwise clean lint diff job.
• gateway/run.py: use the platform-level thread metadata helpers for streaming post-response media delivery, matching the non-streaming path and preserving Telegram DM topic fallback metadata.
• agent/model_metadata.py, agent/models_dev.py: add provider-aware Tencent TokenHub context-length fallback for hy3-preview so direct TokenHub routing does not inherit provider-unaware OpenRouter metadata.
• agent/i18n.py, tests/agent/test_i18n.py: make i18n cache reset safe under patched cache functions and isolate catalog cache state between tests.
• hermes_cli/tools_config.py: infer configured platform toolsets from built-in static toolset members even when registry plugin add-ons are present.
• tests/e2e/conftest.py: bypass destructive slash-command confirmation in e2e slash lifecycle tests, leaving confirmation UX to dedicated tests.
• tests/run_agent/test_async_httpx_del_neuter.py: construct the stale-loop regression cache key through the production helper after pool hints became part of auxiliary client cache keys.

How to Test

Run the focused current-main failed nodes:

./scripts/run_tests.sh -n 0 \
  tests/gateway/test_restart_drain.py::test_restart_command_while_busy_requests_drain_without_interrupt \
  tests/gateway/test_tts_media_routing.py::test_streaming_delivery_routes_telegram_flac_media_tag_to_document_sender \
  tests/gateway/test_tts_media_routing.py::test_streaming_delivery_routes_non_voice_telegram_ogg_media_tag_to_document_sender \
  tests/gateway/test_tts_media_routing.py::test_streaming_delivery_routes_telegram_mp3_media_tag_to_voice_sender \
  tests/run_agent/test_async_httpx_del_neuter.py::TestClientCacheBoundedGrowth::test_same_key_replaces_stale_loop_entry \
  'tests/e2e/test_platform_commands.py::TestSlashCommands::test_new_resets_session[telegram]' \
  'tests/e2e/test_platform_commands.py::TestSlashCommands::test_new_resets_session[discord]' \
  'tests/e2e/test_platform_commands.py::TestSlashCommands::test_new_resets_session[slack]' \
  'tests/e2e/test_platform_commands.py::TestSessionLifecycle::test_new_then_status_reflects_reset[telegram]' \
  'tests/e2e/test_platform_commands.py::TestSessionLifecycle::test_new_then_status_reflects_reset[slack]' \
  'tests/e2e/test_platform_commands.py::TestSessionLifecycle::test_new_then_status_reflects_reset[discord]' \
  'tests/e2e/test_platform_commands.py::TestSessionLifecycle::test_new_is_idempotent[telegram]' \
  'tests/e2e/test_platform_commands.py::TestSessionLifecycle::test_new_is_idempotent[discord]' \
  'tests/e2e/test_platform_commands.py::TestSessionLifecycle::test_new_is_idempotent[slack]' \
  -q --tb=short

Run related regression coverage and blocking static checks:

./scripts/run_tests.sh -n 0 \
  tests/agent/test_i18n.py \
  tests/gateway/test_base_topic_sessions.py \
  tests/gateway/test_tts_media_routing.py \
  tests/hermes_cli/test_tencent_tokenhub_provider.py::TestTencentTokenhubContextLength \
  tests/hermes_cli/test_tools_config.py::test_get_platform_tools_recovers_non_configurable_toolsets_from_composite \
  tests/run_agent/test_async_httpx_del_neuter.py::TestClientCacheBoundedGrowth \
  -q --tb=short

python - <<'PY'
from pathlib import Path
import yaml
yaml.safe_load(Path('.github/workflows/lint.yml').read_text(encoding='utf-8'))
print('lint.yml valid YAML')
PY

uv tool run ruff check .
python scripts/check-windows-footguns.py --all

Validation Status

Local validation on Arch Linux, kernel 7.0.3-arch1-2, Python 3.14.4, branch head 84d4eb74f:

• Current-main failed node set from Tests run 25612702683: 14 passed in 5.02s
• Related regression coverage: 46 passed in 1.02s
• .github/workflows/lint.yml parsed successfully with PyYAML
• uv tool run ruff check .: all checks passed
• python scripts/check-windows-footguns.py --all: no Windows footguns found, 399 files scanned
• Full pytest tests/ -q was not run locally

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
• I've added tests for my changes (required for bug fixes, strongly encouraged for features)
• I've tested on my platform: Arch Linux, Python 3.14.4

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

For New Skills

N/A.

Screenshots / Logs

Focused validation logs are summarized above. CI is running on the updated head and should verify the full affected matrix.