🤖 refactor: move per-provider Coder catalog merge rules into a tested module

[ThomasK33]

Summary

CoderOauthService.refreshBridgeModelsSerializedEffect decides, inside the providers-file lock, what each AI Gateway provider contributes to the next model catalog write. Those per-provider rules (carry-forward, type gating, conclusiveness) and the read of the prior catalog state now live in src/node/services/coderGatewayCatalogMerge.ts, with direct unit tests. Behavior is unchanged.

Background

This PR is part of the CRAP campaign on src/node/services/coderOauthService.ts. CRAP = CC² × (1 − coverage)³ + CC.

Function	CC before	CC after
refreshBridgeModelsSerializedEffect	54	30
reconcileProviderCatalog (new)		18
readPriorCoderCatalog (new)		5
stringEntries (new)		2
idsUnderProvider (new)		1

Scoring method: make crap with the 9 CI crap-lcov-* reports of PR run 38364731596 plus a local lcov of coderOauthService.test.ts and the new test file (bun test --coverage). With it, refreshBridgeModelsSerializedEffect goes from CRAP 54 to 30, and no function in the file is above 30. make crap-check with that coverage: PASS, 0 violations, 1 warning (reconcileProviderCatalog CC 18, move relief: the kept function lost CC 24, the new functions hold 22 decision points).

Note: CI's Quality / CRAP shows this file at about 0% coverage, because scripts/test-unit-ci.sh runs coderOauthService.test.ts as an isolated file without --coverage. The new test file runs in the shared shards, so CI does see its coverage.

Implementation

• readPriorCoderCatalog(section) returns the prior catalog markers (known, discovered, stale, providers). It reads the same fields with the same string filter as before.
• reconcileProviderCatalog(provider, result, prior, authoritative) returns one provider's contribution: the model IDs, the provider metadata to persist on the probe path (zero or one entry), and whether it makes the catalog inconclusive. Each branch and its comment moved verbatim from the loop body.
• The locked mutation maps the results through it and concatenates the IDs and providers in result order, as the loop did. catalogInconclusive is true when any provider is inconclusive. updateProviderSection calls the mutation once, so assigning the flag instead of only setting it to true changes nothing.

Validation

• coderGatewayCatalogMerge.test.ts (16 cases) covers the marker parsing and every ok / error / unavailable branch, on the listing and the probe path, including type changes and the name-as-type default.
• 9 mutants of the moved rules (type gate, name default, stale type gate, stored-metadata preference, prefix match, empty-catalog knowledge, and others) each fail at least one new test.
• coderOauthService.test.ts passes unchanged (101 tests).

Risks

Low. The change is a pure move of catalog merge rules behind the same lock. A mistake would show as a wrong catalog after a refresh (models missing or carried forward wrongly). The existing service tests for those flows pass unchanged.

---

Generated with xum • Model: anthropic:claude-opus-5-5 • Thinking: high

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-13T08:22:04.694688Z	9e84555	PR opened
🔒 Security Review	✅ Completed 2026-10-13T08:23:29.007412Z	9e84555	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.