refactor(plugins): share API provider specs

[steipete]

Six bundled providers now share PluginProviderSpec registration: Hugging Face, Nous, Fireworks, xAI, Venice, and Zed. Hugging Face and xAI also share the API-key settings implementation. This removes 255 production lines and raises the declarative count from 31 to 37, with the 87-provider registry unchanged.

Native boundaries remain provider-owned: Hugging Face keeps its CLI-token reader and serialized retained runtime; Nous keeps Hermes credential validation and diagnostics; Fireworks keeps slug projection and allowlisted persistence; Venice and Zed keep their native source routing, cookie policy, bindings, and visibility. Bundled scripts are unchanged. Every migrated color uses ProviderColor(hex:).

The new typed options carry existing status-page, token-cost, settings-section, and plugin-result-policy contracts. Each is needed by at least two candidates, including deferred providers: status pages by Hugging Face/OpenAI; token costs by xAI/OpenRouter/OpenAI; settings sections by Fireworks/Moonshot/z.ai/OpenRouter; result policies by Fireworks/OpenAI. Shared code contains no provider switch.

Equivalence and validation

The separate pre/post capture is byte-identical: 53,692 bytes, SHA-256:

18e4f2007980d5a6cc990a51e0f9f60f19b791c225c8b04e035f40ce20e96798

It covers all ten candidates: registry and implementation order, full CLI help, metadata/branding/widget flags, settings text/actions/pickers, cookie-mode snapshots, credential projections, token-account metadata, and availability. The committed golden remains derived-only and adds the token-cost capability. New coverage checks native file-token discovery and reuse of the retained Hugging Face strategy.

All Swift tests used source Scripts/test_environment.sh. No live probes, cookie imports, app relaunches, release builds, or full sharded suite were run.

swift test --jobs 4 --filter 'PluginProviderSpecTests|PluginCookieProviderSpecTests|ProviderArchitectureGatekeeperTests|ProviderSettingsDescriptorTests|ProviderSettingsIntegrationTests|ProviderPlugin.*ParityTests|CLIEntryTests|CLICommonOptionsTests|CLIArgumentParsingTests|CLIProviderSelectionTests|ProviderSelectionFixtureTests|CLIUnificationGoldenTests|TokenAccount.*Tests|ProviderTokenAccountData|FileTokenAccountStore|ProviderCredentialCharacterizationTests|ProviderConfigEnvironmentTests|ProviderTokenResolverTests|ProviderLabelMetadataCharacterizationTests|ProviderPresentationPolicyCharacterizationTests|ProviderRegistryTests|ProviderWidgetAvailabilityTests|WidgetProviderChoiceTests|ProviderIconResourcesTests|ProviderAccentColorTests|ConfigurationDocsProviderIDTests|ProviderEndpointOverrideSecurity.*Tests|HuggingFace|NousPluginTests|NousSettingsReaderTests|XAIProviderTests|XAICostUsageMappingTests|Fireworks|Venice|Zed|HostResultProviderTests|ProviderPluginResultTests'

✔ Test run with 565 tests in 60 suites passed after 27.534 seconds.
✔ Test run with 56 tests in 5 suites passed after 4.176 seconds.
✔ Test run with 18 tests in 2 suites passed after 0.041 seconds.

565 app/core tests, 56 plugin tests, and 18 Linux-target tests passed. XCTest executed 65 tests with two opt-in renderers skipped and zero failures. Linux-target tests ran on macOS.

CODEXBAR_PLUGIN_ENGINE=jsc swift test --skip-build --filter 'PluginProviderSpecTests|ProviderPluginParityTests|ProviderPluginDetailsParityTests|HuggingFacePluginTests|HuggingFaceWalletPluginTests|NousPluginTests|XAIProviderTests|FireworksUsageFetcherTests|VeniceUsageFetcherTests|VeniceWebUsageFetcherTests|VeniceClerkSessionTests|ZedPluginTests|ZedStatusProbeTests|HostResultProviderTests'
make check

✔ Test run with 117 tests in 11 suites passed after 2.071 seconds.
✔ Test run with 39 tests in 3 suites passed after 4.225 seconds.
Done linting! Found 0 violations, 0 serious in 2660 files.

JSC: 117 app/core and 39 plugin tests passed. make check: exit 0, zero violations in 2,660 Swift files. All three pre-commit independent reviews finished without actionable P0–P2 findings.

The initial build caught a test naming the removed xAI implementation. A temporary capture hook also reused the golden's fixture name and contaminated synthetic account availability; the hook was removed and the focused run passed without changing the frozen expected values. One argument-layout lint violation was fixed in a separate commit. No pre-existing bug fix or regression red→green claim is made; this is a behavior-preserving refactor. No changelog entry is needed.

Deferred candidates

• Moonshot: provider-specific config normalization (Sources/CodexBarCore/Providers/Moonshot/MoonshotProviderDescriptor.swift:88) and shared Kimi branding remain outside the spec.
• z.ai: custom monthly-MCP pacing (Sources/CodexBarCore/Providers/Zai/ZaiProviderDescriptor.swift:95) is not expressible by the spec; it is the only candidate needing that option.
• OpenRouter: its credit capability (Sources/CodexBarCore/Providers/OpenRouter/OpenRouterProviderDescriptor.swift:59) is unique among these spec candidates; widget selectability also must be retained.
• OpenAI API: the exact (0.06, 0.51, 0.43) color (Sources/CodexBarCore/Providers/OpenAI/OpenAIAPIProviderDescriptor.swift:45) cannot be represented by six-digit hex without changing the capture; its custom icon and cost-menu configuration also remain outside the current spec. No color was changed.

Abacus, Muse, LongCat, Replicate, and TypeSafe are untouched as requested.

Production delta

Group	Insertions	Deletions	Net
API registration/settings	139	351	-212
Venice/Zed registration	81	124	-43
Whole branch	220	475	-255

The branch changes 12 production files. Scripts/regenerate-provider-manifests.sh regenerated the manifests; only the two app implementation entries changed. No gatekeeper anchors needed changes. Tests add 156 lines and remove four relative to the pinned base bf34fc9f424f.

Refs #4043
Refs #4049
Refs #4058

[clawsweeper]

🦞👀
ClawSweeper picked this up.

Pull request received. I will update this pull request when review starts.

ClawSweeper review blocked

Automated review did not run, so no review verdict was produced.

Reason: The input-safety check rejected material in this revision. No detected value, path, or scanner output is reproduced here.

Automatic review is on hold, including after source changes. Request a fresh re-review after resolving the failure; maintainers can also release the hold through an intentional scanner-policy update.

Next step: If this is a genuine credential, remove and rotate it. If it is an intentional test fixture, a maintainer must review and qualify it.

View the workflow run.