[chrisohearn]

The proxy-backed evals (mcp-tool-use with Codex models, and test-signal) crashed with a Python KeyError: 'vibeproxy' traceback when ~/.codex/config.toml had no vibeproxy provider. The traceback didn't say what was missing or that evals are optional.

Fix

• evals/shared/proxy.ts: the TOML reader exits with a sentinel code when the config file, the model_providers table, or the vibeproxy entry is missing. readProxyConfig turns that into a ProxyConfigMissing error. proxyConfig(), which both evals call before any model request, prints that error's message and exits 2, without a stack trace. The Codex home is read from $CODEX_HOME when set, otherwise ~/.codex.
• evals/shared/proxy.test.ts: covers the three missing cases and a configured provider (the control case).
• FORK.md (new "Evals" section) and evals/README.md: on this fork, hosted CI, the Agent context budget check and the review of record gate model-facing text changes. Paid eval runs need the owner's go-ahead for each run.

No new feature. No eval or model call was run.

Acceptance checks

• With no vibeproxy provider configured, node test-signal/run.ts --out <tmp> prints the following and exits 2, without creating any output:

  No `[model_providers.vibeproxy]` in <home>/.codex/config.toml.
  This eval sends model calls through a CLIProxyAPI provider with that name; set one up, or set CODEX_HOME to a Codex home that has it.
  Evals are optional and call paid models. CI does not run them. See evals/README.md.
  

• node mcp-tool-use/run.ts --model gpt-6-astra --case review-round-uses-thread-send --samples 1 --out <tmp> prints the same message and exits 2 before any model call.
• vp test run --config ../vite.config.ts shared/proxy.test.ts --maxWorkers=2 --no-file-parallelism (from evals/): 4 passed.
• effect-tsgo tsrs --noEmit in evals/: exit 0, no errors.
• vp fmt --check and vp lint on the changed files, plus the pre-commit hook's fmt, lint and knip:check: pass.
• Hosted CI green on this head.

The broad local check:changed was not run: on the shared host it ran out of memory. CI runs the full suites.

Model and harness: Claude Opus 5.5 in Claude Code.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary: ✅ Completed

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

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-14T13:28:03.732249Z	fc3671d	PR opened
🔒 Security Review	✅ Completed 2026-11-14T13:29:43.091255Z	fc3671d	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.