feat(models-info): let the modelsInfoUrl catalog hide text-only / unmatched models

[stephane-segning]

1. Summary

This PR changes:

• Adds a new opt-in meta.modelsInfoHideTextOnly provider option to @vymalo/opencode-models-info.
• When true, the modelsInfoUrl catalog becomes authoritative for which models exist, not just their metadata: a model is deleted from provider.models outright (instead of being enriched) if the catalog reports it as text-in/text-out only, or if the catalog has no entry for it at all.
• Updates docs (docs/models-info.md, packages/opencode-models-info/README.md, CLAUDE.md/AGENTS.md composition-contract section) and CHANGELOG.md.

It solves:

• A user question in this session: "Do we have a way of hiding a text model when using model info?" — followed up with "I want the model info endpoint [to] take total precedence over the /models endpoint. Is it possible?" No existing mechanism let the richer modelsInfoUrl catalog prune or override membership; the plugin only ever merged metadata onto models that already existed.

---

2. Intent

The intent of this PR is:

Let a richer, OpenRouter-shaped metadata endpoint be the single source of truth for which models are offered — e.g. hiding chat-only models when only multimodal ones should appear in OpenCode's picker, or overriding a provider's own /v1/models discovery (used by @vymalo/opencode-oauth2) entirely. See docs/models-info.md#hiding-text-only-models--catalog-authoritative-membership for the full design writeup (source of truth for this change).

---

3. Scope

In Scope

• New meta.modelsInfoHideTextOnly boolean option (default false), parsed in config.ts.
• isTextOnlyModality() helper in mapping.ts.
• Deletion logic in plugin.ts's enrichProvider: drops a matched-but-text-only model, and (when the flag is on) drops any model the catalog doesn't mention at all.
• New log events models_info_model_hidden_text_only and models_info_model_hidden_unmatched (debug), plus hiddenCount on the models_info_enriched summary event.
• Tests (config.test.ts, mapping.test.ts, plugin.test.ts) and docs/README/CHANGELOG updates.

Out of Scope

• Synthesizing new provider.models entries for catalog models that discovery/config never listed (only pruning existing entries is supported — see the "Config shape" discussion in the PR description below).
• Any change to the default (flag-off) enrichment behavior — fully backward compatible.

---

4. Verification

I verified this change by:

• Running automated tests
• Checking logs
• Running manual tests
• Checking metrics
• Testing error cases
• Testing permissions/security behavior
• Testing rollback or failure behavior, if relevant

Commands run:

pnpm --filter @vymalo/opencode-models-info test
pnpm -r build
pnpm -r typecheck
pnpm coverage
pnpm lint
pnpm format:check

Results:

@vymalo/opencode-models-info: Test Files 6 passed (6), Tests 78 passed (78)
pnpm -r build: all 10 buildable packages/apps succeeded
pnpm -r typecheck: all packages succeeded
pnpm coverage: all packages passed their per-package thresholds (models-info: 82.59% stmts / 79.57% branches)
pnpm lint / pnpm format:check: no issues (239 files checked)

---

5. Screenshots / Evidence

Add evidence here:

• Screenshot: N/A (backend plugin logic, no UI)
• Logs: new models_info_model_hidden_text_only / models_info_model_hidden_unmatched debug events, exercised in plugin.test.ts
• Metrics: N/A
• Recording: N/A

---

6. Risk Assessment

Risk level:

• Low

Potential risks:

• A user enabling modelsInfoHideTextOnly without realizing "unmatched → dropped" is part of the semantics could lose a hand-configured model that the metadata endpoint simply doesn't know about.
• The flag makes deletion a hard, silent-by-default (debug-level log only) operation — no dry-run mode.

Mitigation:

• Opt-in, off by default; fully documented (README "Hiding text-only models" section spells out both deletion cases explicitly) and covered by tests for both the "flag off" and "flag on" paths.
• Deletions are logged at debug (models_info_model_hidden_text_only / models_info_model_hidden_unmatched) and rolled up into hiddenCount on the models_info_enriched summary event, so the behavior is diagnosable via --log-level DEBUG.

---

7. AI Usage Declaration

AI was used for:

• Understanding existing code
• Generating code
• Generating tests
• Drafting documentation
• Refactoring
• Reviewing the diff
• Not used

Human verification:

• I understand every meaningful change in this PR
• I checked generated code manually
• I checked generated tests manually
• I removed unsupported AI assumptions
• I accept responsibility for this PR

---

8. Reviewer Focus

Please focus your review on:

• Correctness
• Architecture
• Edge cases
• Security
• Performance
• Tests
• Maintainability
• Product intent

[changeset-bot]

⚠️ No Changeset found

Latest commit: a7cb39d

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

[github-actions]

✅ AI Governance check passed

This PR declares AI usage, references a source of truth, and provides verification evidence. Thank you.