Validation / CI note:

Local validation for the actual change set is green:

• python3 -m py_compile agent/models_dev.py hermes_cli/config.py hermes_cli/model_switch.py run_agent.py tests/agent/test_models_dev.py tests/hermes_cli/test_config_validation.py tests/hermes_cli/test_model_switch_custom_providers.py tests/run_agent/test_run_agent.py
• uv run pytest tests/agent/test_models_dev.py tests/hermes_cli/test_config_validation.py tests/hermes_cli/test_model_switch_custom_providers.py -o addopts='' -q
• uv run pytest tests/run_agent/test_run_agent.py::TestBuildApiKwargs::test_reasoning_sent_for_custom_provider_capability_override tests/run_agent/test_run_agent.py::TestBuildApiKwargs::test_reasoning_override_matches_custom_provider_base_url_not_name -o addopts='' -q

Current CI red does not appear to be evidence that this PR regressed the repo:

1. test

• The failure list is broad and mostly outside this diff (memory, gateway session race guards, auth/provider gate, voice mode, interrupt plumbing, browser camofox state, etc.).
• The same clusters are already failing on main push runs as well (for example 24339048768).
• This PR's failures do not cluster around the custom-provider capability override code path.

2. build-and-push

• Fails with the same Docker/npm error already seen on main:
  • npm ERR! syscall spawn git
  • ENOENT
• That's the same class of Dockerfile failure we already observed on current mainline runs.

3. docs-site-checks

• The diagram linter is failing on unrelated docs files / existing ASCII diagram issues outside this PR's diff.
• The file changed in this PR (website/docs/integrations/providers.md) is reported as clean by the same job.

So at this point the red checks look like broad-repo / operational noise, not a targeted regression from this custom-provider capability patch.

Refresh rebuilt from origin/main and pushed to refresh/8942. Validation: py_compile on changed .py files passed; focused pytest on changed test files ran and hit one existing failure in tests/run_agent/test_run_agent.py::TestStreamingApiCall::test_tool_call_accumulation (expected web_search, got search).

Additional validation note:

The one extra focused failure I hit while checking this refreshed branch is reproducible on clean origin/main too:

• /home/ubuntu/hermes-all-venv/bin/python -m pytest -o addopts= -q tests/run_agent/test_run_agent.py::TestStreamingApiCall::test_tool_call_accumulation\n- current origin/main fails with the same "search" vs expected "web_search" assertion\n\nSo that remaining failure is baseline repo noise in current main, not a regression unique to the custom-provider capability override refresh.

✅ Audit completed (2026-04-19)

• Branch refreshed from current origin/main (SHA: 6af04474)
• Validation: Focused tests passing (capability overrides)
• Impact: Medium (allows manual provider capability overrides)
• Risk: Medium (484 lines, 9 files, feature addition)
• Age: Very old PR (5/5)

Baseline Noise Note: Test failures reproduce on clean origin/main — not a regression from this PR.

Recommendation: DECISION NEEDED — Feature adds custom provider capability overrides.

Question for maintainers: Is manual capability override still needed?

• If YES → merge after review
• If NO → close as deprecated

Feature value: Allows users to manually override provider capabilities (e.g., force a model to support vision or tools).

Part of batch audit: 23 PRs audited, 2 closed (absorbed), 21 refreshed
Batch 3 — Feature decision required

🔔 Ready for maintainer review

Esta PR foi validada como parte da auditoria completa de 2026-04-19.

Status:

• ✅ Refreshed from origin/main (SHA: 6af04474)
• ✅ Focused tests passing (or unit tests for fix(firecrawl): support self-hosted browser configuration #6394)
• ✅ Comentários de validação adicionados
• ✅ Baseline noise identified (where applicable)

Ação necessária: Review e merge (ou decisão de feature para #9209, #8942).

Audit batch 3: 4 PRs com validação mínima concluída

Audit/update 2026-04-25:

• Force-pushed d0fe8aaf -> 161594ca with the minimal get_model_info() fix and canonical author metadata.
• Focused local validation: 358 passed in 26.55s after minimal get_model_info() fix.
• Recommendation: Applied minimal custom-provider get_model_info() fix and canonical author rewrite; ready for re-review after CI.

This was part of the open-PR cleanup pass against current upstream/main.

CI follow-up from the 2026-04-25 cleanup pass:

• Contributor Attribution now passes on the rewritten head (161594ca).
• Remaining CI is red: Supply Chain Audit, test, and e2e.
• I reproduced the supply-chain scanner trigger locally. It is firing on stale-branch diff noise (hermes_cli/setup.py and skills/productivity/google-workspace/scripts/setup.py) because the PR diff against current main spans ~490 files.
• The minimal get_model_info() fix itself was validated locally with the focused suite: tests/agent/test_models_dev.py, tests/hermes_cli/test_config_validation.py, tests/hermes_cli/test_model_switch_custom_providers.py, and tests/run_agent/test_run_agent.py — 358 passed.

So this PR is not merge-ready as-is despite the minimal fix and attribution pass. Next step would be a clean rebase/reconstruction onto current main so CI evaluates only the custom-provider capability override change.

Refreshed this PR onto current main with a normal non-force push.

Current PR scope remains limited to custom-provider capability overrides:

• agent/models_dev.py
• hermes_cli/config.py
• hermes_cli/model_switch.py
• run_agent.py
• targeted tests and provider docs

Validation:

• python3 -m py_compile agent/models_dev.py hermes_cli/config.py hermes_cli/model_switch.py run_agent.py tests/agent/test_models_dev.py tests/hermes_cli/test_config_validation.py tests/hermes_cli/test_model_switch_custom_providers.py tests/run_agent/test_run_agent.py -> passed
• HERMES_TEST_WORKERS=2 scripts/run_tests.sh tests/agent/test_models_dev.py tests/hermes_cli/test_config_validation.py tests/hermes_cli/test_model_switch_custom_providers.py tests/run_agent/test_run_agent.py::TestBuildApiKwargs::test_reasoning_sent_for_custom_provider_capability_override tests/run_agent/test_run_agent.py::TestBuildApiKwargs::test_reasoning_override_matches_custom_provider_base_url_not_name -q -> 62 passed
• Post-push PR diff file list verified: 9 scoped files only
• Attribution, docs, check, e2e, Nix, and supply-chain checks are green

The remaining broad Tests workflow is still red. Its failure profile matches current main closely; the one PR-only failure observed (test_live_cap_applied_after_read_add) is outside this PR's custom-provider scope and passes locally in the refreshed worktree.

Closing this for now. Provider capability overrides are a broader config/model-switching feature, and this branch now conflicts in config/model-switch tests. It should be rebuilt from current main if still desired.