ci: unblock fork CI runner selection and npm staging

[dzianisv]

Summary

• resolve npm staging rust-release lookup against openai/codex
• fix the repo-level Prettier failures surfaced by build-test
• keep openai/codex on larger/custom runners while routing forks to hosted runners for rust-ci and Bazel macOS jobs

Why

Two separate CI regressions were blocking real signal on this fork:

1. build-test could not find upstream rust-release runs because the staging script searched the fork by default.
2. rust-ci and Bazel macOS jobs targeted runner labels the fork cannot provision, so they failed before step 1.

In practice issue #27 cannot merge cleanly without issue #28, so this PR is the combined merge vehicle.

Testing

• python3 -m unittest scripts/test_stage_npm_packages.py
• pnpm dlx prettier@3.5.3 --check *.json *.md docs/*.md .github/workflows/*.yml **/*.js
• Ruby YAML parse of .github/workflows/rust-ci.yml and .github/workflows/bazel.yml
• go run github.com/rhysd/actionlint/cmd/actionlint@latest .github/workflows/rust-ci.yml .github/workflows/bazel.yml
• GitHub Actions verification on this PR that the former zero-step jobs now schedule on hosted labels and execute real steps

Issues

• refs fix build-test when staged npm package release workflow is missing #27
• refs investigate rust-ci matrix jobs that fail before any steps run #28

[OpenCodeEngineer]

Pushed follow-up commit d3d11f1137fab1aabd5a9c70db8a5bf808fe4b5d to keep PR #32 moving after the runner fix exposed real hosted-CI failures.

Included in this push:

• cli/build.rs: remove expect() on CARGO_MANIFEST_DIR, fall back to CODEX_CLI_VERSION=unknown
• core/src/codex.rs: remove expect() in the reflection path
• core/src/models_manager/manager.rs: fix newly active clippy lints (redundant_closure_for_method_calls)
• bazel.yml: move fork-vs-upstream matrix selection into fromJSON(...) include lists so forks only schedule currently viable hosted Bazel lanes

Local validation for this SHA:

• just fmt
• cargo test -p codex-cli
• cargo clippy -p codex-cli --all-features --tests -- -D warnings

Current watcher snapshot on this SHA:

• mergeable: yes
• review feedback: none
• checks: 1 passed, 26 pending, 0 failed

I’m continuing to watch this PR until the new SHA settles.

[OpenCodeEngineer]

Pushed follow-up commit 0f01a92c8 after diagnosing the red clippy lanes on the previous SHA.

Root cause of the failures on d3d11f1137fab1aabd5a9c70db8a5bf808fe4b5d:

• all completed Lint/Build failures were the same clippy::single_component_path_imports error on codex-rs/core/tests/common/lib.rs
• the reported Tests — ubuntu-24.04-arm - aarch64-unknown-linux-gnu failure was only a cancellation caused by the workflow already being red, not a distinct test regression

Fix in this commit:

• remove the redundant use serde_json; import from codex-rs/core/tests/common/lib.rs

Local validation:

• just fmt
• cargo clippy -p core_test_support --all-features --tests -- -D warnings

I am restarting the watch on the new SHA now.

[OpenCodeEngineer]

Pushed follow-up commit 365c5a7c6 to clear the workspace test-target clippy failures that appeared after the second hosted rerun.

What this commit fixes:

• codex-rs/tui/tests/suite/model_switching_e2e.rs
  • narrow too_many_arguments expectation on the helper with many test knobs
  • replace the complex return type with a type alias
  • remove the redundant provider_api.clone()
• codex-rs/core/tests/suite/cli_stream.rs
  • remove a needless_collect in the /responses assertion
• codex-rs/core/tests/suite/eval_swe_bench.rs
  • narrow expect_used expectations to the intentionally panic-on-misconfigured Azure eval helpers
  • switch deprecated Command::cargo_bin("codex") to Command::new(codex_utils_cargo_bin::cargo_bin("codex").unwrap())
  • drop the now-unused assert_cmd::prelude::* import
• codex-rs/core/tests/suite/reflection.rs
  • narrow expect_used expectations to the intentionally panic-on-misconfigured Azure reflection helpers
  • switch deprecated Command::cargo_bin("codex") to Command::new(codex_utils_cargo_bin::cargo_bin("codex").unwrap())

Local validation for this batch:

• just fmt
• cargo clippy -p codex-tui --all-features --tests -- -D warnings
• cargo clippy -p codex-core --all-features --tests -- -D warnings with CARGO_BUILD_JOBS=1

I’m resetting the PR-32 watch to this new SHA now. The Linux GNU Bazel fetch failure is still tracked separately as the remaining known flaky/external class.

[OpenCodeEngineer]

Pushed follow-up commit e607fb8e4 to address the real hosted failure that surfaced on the previous SHA.

Root cause from the Bazel log on 365c5a7c6:

• core/tests/common/lib.rs tried to include_str!("../../models.json")
• that relative compile-time path does not exist when core/tests/common is built as its own Bazel package
• this broke both the Bazel workflow and hosted test lanes that compile core_test_support

Fix in e607fb8e4:

• replace the compile-time include with a shared runtime loader using codex_utils_cargo_bin::find_resource!("../../models.json")
• reuse the same loader from core/tests/common/test_codex.rs
• remove the now-unused anyhow::Context import

Local validation on the final tree:

• cargo test -p core_test_support ✅
• bazel test //codex-rs/core/tests/common:common-unit-tests --test_output=errors ✅

Watcher is now following the new head SHA e607fb8e4e47353c0009486bab0eca00b0762cb0, which has already restarted CI.

[OpenCodeEngineer]

Pushed 1a8218d135a642244c182a77805fb7b32757bc83 to this PR (codex: fix PR #32 hosted test regressions).

This follow-up addresses the hosted failures that were still exposed after the runner fallback work:

• app-server and core test cache fixtures now include provider_scope, matching the current models_cache.json schema
• TUI model-switch tests now avoid in-test Cargo builds by preferring sibling/target binaries, and Windows TOML fixture paths are quoted correctly
• rust-ci now prebuilds codex and test_stdio_server before nextest
• memories prompts now use Bazel-safe include_str!() templates instead of Askama file path resolution
• the experimental popup snapshot is split for Linux so the Bubblewrap row does not break the non-Linux snapshot

Local validation on this SHA:

• cargo test -p core_test_support
• cargo test -p codex-core --test all suite::models_cache_ttl::uses_cache_when_version_matches -- --nocapture --exact
• cargo test -p codex-core --test all suite::remote_models::remote_models_truncation_policy_with_tool_output_override -- --nocapture --exact
• cargo test -p codex-core --test all suite::rmcp_client::stdio_image_responses_are_sanitized_for_text_only_model -- --nocapture --exact
• cargo test -p codex-core memories::prompts::tests -- --nocapture
• cargo test -p codex-app-server --test all suite::v2::model_list -- --nocapture
• cargo test -p codex-tui --test all suite::model_switching_e2e::models_dev_provider_config_parses_custom_provider -- --nocapture --exact
• cargo test -p codex-tui --test all suite::model_switching_e2e::ollama_model_switch_then_prompt_uses_responses_api -- --nocapture --exact
• cargo test -p codex-tui --test all suite::model_availability_nux::resume_startup_does_not_consume_model_availability_nux_count -- --nocapture --exact
• cargo test -p codex-tui chatwidget::tests::experimental_popup_marks_reflection_enabled_when_feature_flag_is_enabled -- --nocapture

GitHub reruns are active on this commit now. Current check snapshot immediately after push: 4 passed, 23 pending, 0 failed. I’m keeping watch on the new SHA.

[OpenCodeEngineer]

Pushed a narrow workflow-only follow-up at d6a06413f38b688f4529aebbd6c14e110a57a420 (codex: fix PR #32 windows arm node setup).

Why this was needed:

• the rerun on 1a8218d13 exposed a new immediate failure in Tests — windows-11-arm - aarch64-pc-windows-msvc
• the job never reached Rust; it failed in step 3, actions/setup-node@v6, immediately after resolving codex-rs/node-version.txt to Node 22.22.0

What changed:

• keep the normal Node setup path for every non-Windows-ARM test lane
• install x64 Node explicitly on the Windows ARM test lane so js_repl setup uses the x64 runtime under emulation instead of the failing default architecture path

Validation:

• local YAML parse of .github/workflows/rust-ci.yml

The matrix restarted on d6a06413f and is pending again. Current snapshot right after push: 2 passed, 25 pending, 0 failed. I’m keeping watch on the new rerun.

[dzianisv]

Superseded by #35, which now carries this branch's changes on the consolidated integration branch.