Skip to content

Add OpenAI 6.2 models without pinning provider catalogs - #32

Merged
mergify[bot] merged 2 commits into
mainfrom
HOUSE-131-lane
Nov 17, 2026
Merged

mergify[bot] merged 2 commits into
mainfrom
HOUSE-131-lane

Conversation

@ericlitman

@ericlitman ericlitman commented Nov 17, 2026 •

Copy link
Copy Markdown
Contributor

Summary

Keep OpenAI 6.2 discovery without pinning complete provider catalogs.

public catalog refresh
  keep every upstream entry
  append missing paid Codex 6.2 definitions and native client records
  publish the catalog

Upstream entries take precedence independently by model ID. Free plans are unchanged. This permits removing the studio2 full-catalog overrides that currently stop unrelated model updates.

Evidence

  • Before: both new regression tests fail on unchanged main c7430f1: paid models are missing and the client snapshot lacks native records.
    After: both pass on this branch. The tests also prove unrelated Claude and older Codex updates propagate, upstream Sol supersedes only its fallback, native records remain intact, and identical reloads do not churn revisions.
  • Parent go test ./... and server compile pass on committed head a52b7d6. The implementation lane observed an existing intermittent heartbeat test failure; parent full-suite run passed.
  • Parent launched the actual server on loopback with synthetic discovery-only credentials, normal public sources and no catalog overrides. Both IDs appeared through /v1/models and the Codex model endpoint; native instructions were retained. Codex response: 782,287 bytes for that isolated credential set.
  • Existing broader response-size regression including both new native templates: 1,024,168 bytes, below 1 MiB. Deployed full catalog: 982,578 bytes.
  • Generated fallback entries deep-equal the previously fetched OpenAI native records and staged paid metadata. Native instruction strings are unchanged.
  • Proof record: projects/cliproxyapi/records/2026-11-17-batch-HOUSE-131/ in the docs repository.

Merge Danger

Door: two-way

The previous binary and private config backup remain available. The full-catalog overrides were removed after deployment and live verification.

Blast Radius: catalogs

Paid Codex registration and client snapshot contents change. Remote refresh code and other provider models are unchanged. The supplemental general entries inherit the predecessor's unverified 128,000 output-token metadata; future upstream entries supersede it.

Acceptance criteria

  • “Add gpt-6.2-sol and gpt-6.2-astra without replacing or pinning the public general or Codex client catalog. Preserve all existing model entries and native 6.2 metadata.” Verified by regression tests, data equality and actual loopback endpoints without overrides.
  • “Remote catalog changes for unrelated providers and older Codex models continue to be applied. If upstream later supplies a 6.2 entry, prefer that upstream metadata rather than freezing a local duplicate.” Verified by the actual publish/load paths in the two regressions.
  • “Keep free-plan exposure unchanged and preserve paid-plan access, reasoning capabilities and native instructions for the new models.” Paid/free, metadata and native record assertions pass.
  • “Remove the two HOUSE-130 full-catalog overrides from studio2 config after a green, reviewed replacement is deployed. Preserve other settings and private rollback material.” Completed after deployment: both overrides removed, other config preserved, previous binary/private config retained.
  • “Prove live discovery and requests for the two models, T3 discovery on all four hosts, client response below 1 MiB, and a deterministic regression test that remote model changes still propagate with the two additions present.” Regression and premerge server proof complete; both production model requests completed with HTTP 200 and T3 discovery was verified on all four hosts.

Gate results

Parent committed head a52b7d6: targeted regression exit 0; go build -o <scratch>/parent-test-output ./cmd/server exit 0; go test ./... exit 0. Changed Go files are gofmt-clean. Scoped no-comments review: zero deletions or flags; only required go:embed directives were added.

Not verified

Deployed as v8.0.44-mob.7 on studio2; both full-catalog overrides removed; both real /v1/responses requests succeeded and four-host T3 discovery passed. Web search, experimental tools and maximum output length are outside this model-discovery proof. Supplemental records remain dormant when upstream supplies the model; retirement behavior is unchanged from existing local model supplements.

Linear: HOUSE-131

Merged review dispositions

Unfret returned request_changes. All three published mandatory findings were classified under the standing consequence/reachability policy without a fix push:

  • Catalog above 1 MiB with 120 compatibility models: pre-accepted, disposition 13105. Actual deployment has no compatibility providers and serves 60 models in 982,578 bytes.
  • Test error-variable shadowing: inert, disposition 13106; no runtime or test behavior consequence.
  • Missing capability metadata for a Codex API-key upstream without an explicit model list: pre-accepted, disposition 13107. Actual deployment has no Codex API-key upstreams; OAuth capability lookup uses the supplemented paid getters.

These are recorded limitations, not code fixes or a green Unfret verdict. No non-Unfret review threads were present.

Eric A. Litman and others added 2 commits November 17, 2026 14:35
@ericlitman
ericlitman marked this pull request as ready for review November 17, 2026 19:42
@mergify
mergify Bot merged commit 5553e76 into main Nov 17, 2026
10 of 13 checks passed
@mergify

mergify Bot commented Nov 17, 2026

Copy link
Copy Markdown
Contributor

✅ Merged into main as 5553e76 at 19:53 UTC

Less than a minute in the queue, with no time running CI.

So far

  • ✅ Entered the queue Tue 17 Nov, 19:53 · rule default · triggered by merge protections
  • ✅ Checks skipped · PR is already up-to-date
  • ✅ Merged 19:53 · 5553e76
9 merge conditions met
  • github-review-approved · required by GitHub repository ruleset rule Mergify merge protections
  • -draft · required by Merge Protections rule Required checks
  • -label = hold · required by Merge Protections rule Required checks
  • check-success = Analyze (actions) · required by Merge Protections rule Required checks
  • check-success = Analyze (go) · required by Merge Protections rule Required checks
  • check-success = Analyze (javascript-typescript) · required by Merge Protections rule Required checks
  • check-success = e2e-hermetic · required by Merge Protections rule Required checks
  • check-success = go · required by Merge Protections rule Required checks
  • any of · required by GitHub repository ruleset rule Mergify merge protections:
    • check-success = @mergify/Mergify Merge Protections
    • check-neutral = @mergify/Mergify Merge Protections
    • check-skipped = @mergify/Mergify Merge Protections

Times in UTC · dashboard · how the merge queue works

@unfret

unfret Bot commented Nov 17, 2026 •

Copy link
Copy Markdown

🛑 3 blocking findings

🔴 High · Keep capability lookup consistent with paid fallback discovery

🟡 Medium · Keep supplemented catalogs below the Codex response limit

🔵 Low · Avoid shadowing the fixture-read error

Walkthrough

Adds gpt-6.2-sol and gpt-6.2-astra to the paid Codex tiers (team, plus, pro) and to the Codex client model catalog by appending embedded fallback definitions only when the upstream catalog lacks that model ID, so remote catalog refreshes keep flowing and an upstream entry supersedes its local fallback. Free-tier exposure is unchanged, and identical reloads don't bump the client catalog revision. Adds tests for remote-update propagation, upstream precedence, and native-instruction retention. The author reports that the studio2 override removal and production checks will happen after deployment.

Area What changes Review notes
Paid Codex model registry Team, plus, and pro model lists now append cloned 6.2 fallback ModelInfo entries for IDs missing upstream. Capability lookup may not apply the same fallback logic. Keep capability lookup consistent with paid fallback discovery
Codex client catalog Loading the client catalog now merges in native 6.2 records, matched by slug, which preserves top-level fields and upstream entries. The larger response is still under 1 MiB in the existing test fixture. Real catalogs could exceed that size. Keep supplemented catalogs below the Codex response limit
Tests Adds two regression tests that cover fallback/upstream interaction, and extends the response-size test to include the new models. A fixture-read error variable is shadowed. Avoid shadowing the fixture-read error
Since the last review
  • New: Keep supplemented catalogs below the Codex response limit
  • New: Avoid shadowing the fixture-read error
  • New: Keep capability lookup consistent with paid fallback discovery
Repair prompt

Read the Unfret check run for mobilyze-llc/CLIProxyAPI#32 at head a52b7d64cbd7bcb1691229f0f7edccc1411f877f (run panel:a52b7d64cbd7bcb1691229f0f7edccc1411f877f:YjcKAwDkG78:LMyA3ms3P2E, check). Use the check's App id and GET /repos/mobilyze-llc/CLIProxyAPI/check-runs/134315986401; its output.text contains the structured result. Compare it with the current code and head, verify that each listed mandatory finding still applies, repair only those that do, honor the repository constraints, and validate the changes. Leave advisories and accepted risks alone. A manually resolved thread is not a waiver.

Details
Lane / stage Model Effort Outcome Duration
broad-gpt-5.6-sol-xhigh / review openai/gpt-5.6-sol → gpt-5.6-sol · primary xhigh succeeded 2 min 59 s
broad-opus-5.5-max / review anthropic/claude-opus-5-5 → claude-opus-5-5 · primary max succeeded 7 min 6 s
focal-gpt-5.6-sol-xhigh copy / review openai/gpt-6.1-sol → gpt-6.1-sol · primary xhigh succeeded 7 min 40 s
focal-gpt-5.6-sol-xhigh copy 2 / review openai/gpt-6.1-sol → gpt-6.1-sol · primary xhigh succeeded 10 min 26 s
judge / judge openai/gpt-6-astra → gpt-6-astra · primary high succeeded 2 min 20 s

First review of this pull request.
4 reviewers finished.
Commit a52b7d6.
Check run.
Run: GET /unfret/run/panel:a52b7d64cbd7bcb1691229f0f7edccc1411f877f:YjcKAwDkG78:LMyA3ms3P2E.

@unfret review · @unfret status · @unfret help

@unfret unfret Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🛑 3 blocking findings
🔴 High · Keep capability lookup consistent with paid fallback discovery
🟡 Medium · Keep supplemented catalogs below the Codex response limit
🔵 Low · Avoid shadowing the fixture-read error
Check

Comment thread internal/registry/model_definitions.go Resolved
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant