Greptile Summary

This PR replaces the previous run_all_tests.sh startup script with an act-driven local CI entrypoint that mirrors the same GitHub Actions jobs (unit-tests, integration-app-harness, benchmarks, check_latest_node) developers would encounter in CI, and documents the new workflow in CONTRIBUTING.md.

Key points:

• The script is well-structured with clear helper functions and sensible environment-variable overrides for platform images, Python versions, and state manager selection.
• A Bash compatibility bug exists: expanding empty arrays ("${act_env_prefix[@]}", "${extra_act_args[@]}") under set -euo pipefail will abort the script with an unbound variable error on Bash ≤ 4.3 (including macOS's default Bash 3.2). The standard fix is "${arr[@]+"${arr[@]}"}"
• ensure_act checks executability with [[ -x "$ACT_BIN" ]], which only works for absolute paths; a user setting ACT_BIN=act (bare name) gets a misleading "install act" error instead of a path-resolution failure.
• CONTRIBUTING.md lists three scrubbed variables in self-hosted mode but the script actually scrubs four — GITHUB_CODESPACE_TOKEN is omitted from the docs.

Confidence Score: 3/5

• Safe to merge for users on modern Bash, but will silently break on macOS default shell and other Bash < 4.4 environments due to the empty-array expansion bug.
• The bash compatibility issue with empty array expansion under set -u is a real runtime failure on macOS (Bash 3.2) and any system with Bash ≤ 4.3 — a common developer environment. Combined with the documentation inaccuracy around scrubbed variables and the ensure_act edge case, the script needs a small but concrete fix before it can be relied upon across all supported environments.
• scripts/run_all_tests.sh — the empty array expansion pattern and ensure_act logic need attention.

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A([bash scripts/run_all_tests.sh args...]) --> B{-h / --help?}
    B -- yes --> C[show_help & exit 0]
    B -- no --> D[resolve ACT_BIN via command -v act]
    D --> E[ensure_act\nvalidates binary exists & is executable]
    E -- fail --> F[error: install act & exit 1]
    E -- ok --> G{is_list_mode args?}
    G -- yes --> H[skip Docker check]
    G -- no --> I{docker_daemon_available\nOR self_hosted_mode_enabled?}
    I -- no --> J[error: Docker required & exit 1]
    I -- yes --> H
    H --> K[configure_platform_args\nset -P ubuntu-latest= & -P ubuntu-22.04=]
    K --> L[configure_self_hosted_env\noptionally prepend env -u vars]
    L --> M[cd repo_root]
    M --> N[run_job: unit-tests\nunit_tests.yml, ubuntu-latest, python 3.13]
    N --> O[run_job loop split 1 & 2:\nintegration-app-harness\nintegration_app_harness.yml]
    O --> P{REFLEX_ACT_SKIP_BENCHMARKS?}
    P -- no --> Q[run_job: benchmarks\nperformance.yml]
    P -- yes --> R
    Q --> R[run_job loop split 1 & 2:\ncheck_latest_node\ncheck_node_latest.yml]

_{Last reviewed commit: 265101f}

Fair pushback. My intent here is not to replace the normal pytest loop; the direct unit-test command still stays the fast path in CONTRIBUTING. This script is an opt-in way to exercise a representative subset of the existing GitHub Actions jobs locally, since some failures show up in workflow/setup parity rather than in pytest alone.

On the images, agreed they are unofficial. I used them as a pragmatic act default, not because I think that tradeoff should be hidden. If the team would prefer, I can make the platform mapping more explicit/configured instead of defaulting it.

I also pushed 93d81e7 for the three smaller issues Greptile called out: the Bash empty-array compatibility case, bare-name ACT_BIN resolution, and the missing GITHUB_CODESPACE_TOKEN mention in the docs.