fix(catalog): read GitHub Copilot context window metadata

[Ingwannu]

Summary

• read GitHub Copilot's live context-window field at capabilities.limits.max_context_window_tokens
• preserve the existing metadata precedence when both old and Copilot-shaped fields are present
• reject malformed/non-positive values through the existing safe-integer boundary
• add an end-to-end routed catalog regression for accepted, conflicting, and invalid Copilot payloads

Root cause

The live-model parser recognized metadata.limits.max_context_length, metadata.context_length, and top-level compatibility fields, but GitHub Copilot publishes the value under capabilities.limits.max_context_window_tokens. The value was therefore lost before catalog construction and Codex fell back to the conservative 128K window.

Verification

Exact head 486b2f99f3182acf055274755ade9c6571203ac9:

• focused Copilot regression: 1 passed, 0 failed
• bun test tests/codex-catalog.test.ts: 255 passed, 0 failed
• bun run typecheck: this branch and an unmodified dev worktree both report the same three pre-existing fetch(..., { timeout }) type errors in src/server/claude-messages.ts and src/server/responses/fetch-helpers.ts; this diff adds no type errors
• all local runs used isolated HOME, OPENCODEX_HOME, and CODEX_HOME, nice -n 10, and a two-CPU affinity

Full exact-head CI is still required before merge.

Closes #3156.

Summary by CodeRabbit

• Bug Fixes
  • Improved live model discovery for GitHub Copilot by recognizing context-window limits reported through capabilities.
  • Preserved explicitly configured context-window values when they differ from live discovery data.
  • Ignored invalid negative context-window values.

[github-actions]

✅ Deterministic PR hygiene checks passed.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Team

Run ID: 6af26e8b-ef8d-48d7-8555-9791fe549f78

📥 Commits

Reviewing files that changed from the base of the PR and between 7ac4c9f and 486b2f9.

📒 Files selected for processing (2)

• src/codex/catalog/provider-fetch.ts
• tests/codex-catalog.test.ts

Included review availability: Your plan provides up to 10 included reviews per hour; 9 remain after this review.

---

📝 Walkthrough

Walkthrough

The catalog parser reads capabilities.limits.max_context_window_tokens as a fallback context window for Copilot models. Tests verify fallback resolution, metadata precedence, and rejection of negative values.

Changes

Copilot context window discovery

Layer / File(s)	Summary
Context window fallback and validation src/codex/catalog/provider-fetch.ts:1198-1206, tests/codex-catalog.test.ts:5369-5411	catalogHintsFromModelsApiItem uses capabilities.limits.max_context_window_tokens after limits.max_context_length. Tests verify the 1M capability value, 256k metadata precedence, and rejection of a negative value.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk: ⚪ Minimal · up to 486b2

The PR adds support for GitHub Copilot context-window metadata while preserving existing precedence and validation behavior; no actionable merge-blocking risk remains after normal checks and review.

Suggested reviewers: lidge-j

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 1 functions across 2 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly describes the primary change: reading GitHub Copilot context-window metadata in the catalog parser.
Linked Issues check	✅ Passed	The changes satisfy issue #3156. src/codex/catalog/provider-fetch.ts reads capabilities.limits.max_context_window_tokens as a fallback, while preserving metadata.limits.max_context_length precedence a…
Out of Scope Changes check	✅ Passed	The changes are limited to the requested Copilot context-window parsing fix and regression tests. No unrelated code or behavior changes are present in the provided summary.

Full details: Linked Issues check

Explanation

The changes satisfy issue #3156. src/codex/catalog/provider-fetch.ts reads capabilities.limits.max_context_window_tokens as a fallback, while preserving metadata.limits.max_context_length precedence and safe-integer validation. tests/codex-catalog.test.ts covers valid, conflicting, and invalid Copilot payloads.

• Fix all pre-merge checks with AI

✨ Finishing Touches 💡 1

📝 Generate docstrings 💡

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch ingw/fix-copilot-context-3156

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[lidge-jun]

리뷰 · 우선순위 67 / 80

설명

이 PR은 이슈 #3156을 고칩니다. GitHub Copilot 라이브 모델 페이로드가 context window를 capabilities.limits.max_context_window_tokens 에 두는데, OpenCodex 카탈로그 파서가 그 필드를 읽지 않아 Codex /status 가 보수적 128k로 떨어지던 구멍입니다. 패치는 그 필드를 읽고, 예전 metadata 필드와 같이 있으면 기존 우선순위를 유지하며, 비정상·비양수 값은 기존 positiveSafeInteger 경계로 버립니다. 수락·충돌·무효 Copilot 페이로드 회귀를 tests/codex-catalog.test.ts 에 넣습니다.

지금 HEAD src/codex/catalog/provider-fetch.ts 의 catalogHintsFromModelsApiItem 은 metadata.limits.max_context_length, metadata.context_length, top-level context_length/context_size/max_model_len/max_context_length, llama.cpp meta.n_ctx 만 모읍니다. Copilot 경로는 없습니다. #2943/#2944 가 nested supports.vision 은 읽게 했지만 context window 필드명은 다른 축입니다. 이 PR diff는 plainRecord(item.capabilities)?.limits 를 후보에 추가합니다. +49/−0, MERGEABLE, 브랜치 ingw/fix-copilot-context-3156 입니다.

typecheck의 기존 fetch(..., { timeout }) 세 오류는 이 diff와 무관하다고 본문이 적습니다. types.ts/config.ts 분할 캠페인과 무관합니다. remote hub 열차와 파일도 겹치지 않아 독립 랜딩하기 좋습니다. 모델마다 16k~1M 창이 128k로 보이면 사용자가 컨텍스트를 잘못 자릅니다.

점수는 67입니다. 작은 카탈로그 정확도 수정이고 #3156과 1:1입니다.

라인 src/codex/catalog/provider-fetch.ts contextWindow 후보 목록 - Copilot 필드를 기존 metadata 후보보다 뒤에 두어 "둘 다 있으면 예전 필드 우선"이 되어야 본문과 같습니다
라인 tests/codex-catalog.test.ts Copilot 회귀 - routed catalog end-to-end가 실제 catalogHintsFromModelsApiItem 을 타는지 확인하십시오
경로 issue #3156 - 머지 시 이슈를 닫으십시오. Closes 링크를 확인합니다
경로 #2943/#2944 - vision과 한 changelog에 적어도, 이 PR 범위에는 넣지 마십시오
경로 positiveSafeInteger - 0/음수/비숫자는 창을 만들지 않아 128k 폴백이 남습니다. 그게 안전 기본입니다

메인테이너의 판단이 필요한 지점

• capabilities.limits를 전 제공자에 열어 두어도 되는지(양수 정수만 받으므로 대체로 안전)
• 2.40.0 릴리스 노트에 Copilot context window 한 줄을 넣을지

너의 추천
머지하고 #3156을 닫습니다. 작은 독립 버그픽스입니다. 라벨은 바꾸지 않습니다.

이 댓글은 grok-bot이 작성했습니다