fix(model-switch): merge providers.<slug>.models into HERMES_OVERLAYS picker rows

[pierrenode]

What does this PR do?

list_authenticated_providers() (hermes_cli/model_switch.py) merges a providers.<slug>.models config block into a built-in provider's (section 1, PROVIDER_TO_MODELS_DEV) discovered catalog (#63055), so a model declared there shows up in the /model picker. Section 2 (HERMES_OVERLAYS — nous, openai-codex, copilot, opencode-go, etc.) builds its model_ids through the same shape of code but never applies this merge.

Root cause: a model declared under providers.nous.models (or providers.opencode-go.models, providers.openai-codex.models, providers.copilot.models) is already typeable via /model <name> today — _configured_provider_matches() (line 728) scans user_providers generically by slug, with no built-in/overlay distinction, specifically to route a typed model name to the right provider (#45006). But it never appears in the picker list for that provider, because section 2's model_ids construction has no equivalent merge. This is inconsistent with the identical config surface already working correctly for built-in providers.

Fix: apply the same merge (configured models first, deduplicated against the discovered list) right before model_ids is capped to max_models, keyed by hermes_slug — the same slug the row itself reports as "slug" in the picker, and the same namespace providers.nous/providers.openai-codex are documented as (hermes_cli/auth.py's own comments reference these exact keys).

Scope note

Section 2b (CANONICAL_PROVIDERS) has the same "providers.<slug>.models is invisible to the picker" symptom for a different subset of providers. I found and fixed it too, but two other open PRs already actively rework that exact code region with a different approach:

• fix: skip canonical providers when user has config-defined models #21983 (skip canonical providers when user has config-defined models) — a slug present in user_providers at all makes section 2b skip the row entirely and defer to section 3, discarding the discovered catalog.
• fix(model): make picker discovery config-aware #60656 (make picker discovery config-aware) — adds a new _is_configured_builtin_provider() gate at the top of both section 2 and section 2b.

Both touch the same few lines my section-2b fix would have touched, with a philosophy that conflicts with a pure merge (skip-and-defer vs. merge-and-keep-both). Rather than add a third competing change to contested territory, I scoped this PR to section 2 only, where there's no existing open PR.

Related Issue

None filed — found via code review while auditing today's (2026-07-12) #63055/#63052/#63046 model-picker series for consistency across all four sections of list_authenticated_providers().

Type of Change

• 🐛 Bug fix

Changes Made

• hermes_cli/model_switch.py: section 2 (HERMES_OVERLAYS loop) now merges providers.<hermes_slug>.models into model_ids before capping to max_models, mirroring section 1's existing merge (+13 lines)
• tests/hermes_cli/test_configured_builtin_models.py: two new regression tests — test_configured_models_extend_hermes_overlay_provider_row and test_configured_models_are_merged_before_picker_limit_for_overlay, mirroring the existing section-1 tests in the same file but against a synthetic HermesOverlay entry

How to Test

python3.11 -m pytest tests/hermes_cli/test_configured_builtin_models.py tests/hermes_cli/test_overlay_slug_resolution.py tests/hermes_cli/test_model_switch_custom_providers.py tests/hermes_cli/test_opencode_zen_model_limit.py tests/hermes_cli/test_ollama_cloud_provider.py tests/hermes_cli/test_model_switch_filter_unresolved.py tests/hermes_cli/test_openai_picker_curated.py tests/hermes_cli/test_list_picker_providers.py tests/hermes_cli/test_user_providers_model_switch.py tests/hermes_cli/test_codex_cli_model_picker.py tests/hermes_cli/test_bedrock_model_picker.py tests/hermes_cli/test_picker_prewarm.py tests/hermes_cli/test_copilot_in_model_list.py tests/hermes_cli/test_models_dev_preferred_merge.py tests/hermes_cli/test_opencode_go_in_model_list.py tests/hermes_cli/test_authenticated_providers_exhausted_pool.py -v --override-ini="addopts="

199 passed. New tests mutation-verified: stashing the fix reproduces the exact bug — both new tests fail with row["models"] == ["live-a", "shared"] (configured models missing), confirming they genuinely catch the regression.

Checklist

• Contributing Guide read | Conventional Commits | Duplicate PR checked (gh pr list --search "HERMES_OVERLAYS configured models picker" / "opencode-go providers models picker" / "list_authenticated_providers configured models" — found fix: skip canonical providers when user has config-defined models #21983/fix(model): make picker discovery config-aware #60656/fix(vertex): register in PROVIDER_REGISTRY and surface as authenticated in desktop picker #57399, all confirmed scoped to section 2b only, not section 2; see scope note above)
• Only this fix | Tests added | Platform: macOS (pure Python control-flow, no OS-specific code)
• Docs — N/A (no new config key, providers.<slug>.models is already documented) | Cross-platform — N/A

[teknium1]

Thanks for tracing this to the overlay path. Current main still has the gap: section 2 builds the HERMES_OVERLAYS catalog and applies its picker limit without the configured-model merge (hermes_cli/model_switch.py:1855-1922), whereas section 1 already performs that merge before truncation (hermes_cli/model_switch.py:1745-1758). The proposed implementation matches that established precedence and deduplication behavior.

Problems

• The new fixture uses test-overlay for both the overlay key and the configured slug. It does not exercise the key distinction that section 2 explicitly supports: a models.dev key such as github-copilot resolved to the config/Hermes slug copilot (hermes_cli/model_switch.py:1781-1792).

Suggested changes

• Add a mapped-slug regression case: overlay key github-copilot, PROVIDER_TO_MODELS_DEV mapping copilot -> github-copilot, and providers.copilot.models. Assert configured IDs precede and deduplicate discovered IDs before the limit.

Automated hermes-sweeper review.

[GottZ]

This was generated by AI during triage.

Summary

Two PRs address the configured-model picker gap across distinct provider paths. #63055 fixed built-in provider rows, while #64877 extends the same configured-first, deduplicated merge to HERMES_OVERLAYS rows.

Related pull requests

• #63055 [merged] related — (+53/-0) — merged reference implementation: merges providers..models into built-in picker rows before max_models truncation, preserving configured models and deduplicating discovered entries. It remains relevant because #64877 applies this established behavior to the separate overlay path.
• #64877 related — (+103/-0) — keep open pending test correction: the implementation merges providers.<hermes_slug>.models into overlay rows before truncation and correctly keys lookup by the resolved Hermes slug. The contributor keep_open review is not fully addressed: the mapped-slug test uses max_models=None, while the separate limit test uses an unmapped synthetic slug, so no test verifies mapped-slug resolution and configured-model precedence under an active picker limit together.

Suggested consolidation

Do not merge #64877 yet. Update its mapped-slug regression to use a restrictive max_models value and assert that providers.copilot.models entries precede and deduplicate discovered models before truncation; after that contributor review requirement is covered and tests pass, merge #64877. #63055 is complementary merged groundwork, not a duplicate, so no PR should be closed as a duplicate.

Cross-PR triage: Reviewed 2 pull requests and 0 issues in this complex. Each diff was read against this issue; Assessment working set: 9 kB of PR diffs, 6 kB of issue/PR text, 1 kB of discussion (1 comments), 0 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.

[pierrenode]

Rebased onto current `upstream/main` and squashed to a single commit.

`hermes_cli/model_switch.py` auto-merged cleanly. `tests/hermes_cli/test_configured_builtin_models.py` had one conflict: the pre-existing test this PR's diff shows as unchanged context (`test_configured_models_are_merged_before_picker_limit`, section-1/built-in coverage, not part of this PR) has been pruned from `upstream/main` (test-pruning wave). Kept upstream's pruning and restored only this PR's own new additions (the `_overlay_row` helper + all 3 section-2/HERMES_OVERLAYS regression tests, starting from `_overlay_row` onward) — verified against the original commit's diff hunk boundaries to isolate exactly what this PR added.

Targeted suite (4 tests) passes, mutation-verified (removing the `_configured_models` merge makes all 3 overlay tests fail — picker rows come back without the configured models prepended). Broader sweep (`tests/hermes_cli/` filtered to model_switch/picker/configured_builtin, 217 tests) passes clean. Ruff clean.

Disclosure — genuine duplicate found: fresh competitor search turned up #72049 ("Show configured overlay models in the desktop picker", opened 2026-07-26, ~11 days after this PR), which implements the identical fix — same file, same insertion point (right before `total = len(model_ids)` in the HERMES_OVERLAYS section), same mechanism (`configured_models = _declared_model_ids(...)`, `model_ids = list(dict.fromkeys([*configured_models, *model_ids]))`). This PR is the older of the two. I'm not in a position to close another author's PR, so flagging here — one of the two is fully redundant once either merges.