test: cover OpenAI compatible execution validation

[msparsh13]

Summary

Add regression coverage for validation performed by OpenAICompatibleExecution.__post_init__.

The tests now cover previously untested invalid configurations, including model whitespace, invalid response format types, malformed API key environment variable names, invalid max_tokens / max_retries values, boolean integer edge cases, and non-finite temperatures.

Validation

Ran:

• uv run pytest tests/test_build_ai_vlm_checks.py -q
• uv run pytest -q
• uv run ruff check --fix
• uv run ruff format
• uv run ty check

All relevant checks passed.

Checklist

• I added or updated outcome-focused tests for changed business logic.
• I updated documentation for changed behavior, flags, formats, or requirements.
• I ran uv run ruff check --fix, uv run ruff format, and uv run ty check.
• I ran the relevant pytest suite.
• I did not add recordings, generated media, credentials, private URLs, or runtime artifacts.
• I preserved stored-data compatibility or documented an explicit version change.

[github-actions]

👋 Hi @msparsh13 — thank you so much for your first contribution to HFlow!

A maintainer will review your pull request as soon as possible. In the meantime:

• ⭐ Star the repo: https://github.com/Hebbian-Robotics/hflow
• 💬 Join our Discord: https://discord.gg/vacepQvjmg
• 📖 Check CONTRIBUTING.md for the validation commands CI will run
• 📣 Share the project with your network

💡 Tip: one open pull request per contributor at a time. Issues with an assignee are taken; everything else is fair game.

We are excited to have you here and appreciate your help making the project better! 🙌

[Sagar-024]

Hi @msparsh13, welcome and nice work. The new rows read exactly like the validation function's docs.

Two small things before merge:

1. max_tokens has the same bool guard (build_ai_vlm_checks.py:161) but no row with max_tokens=True. Worth adding.
2. Heads up, fix(build-ai): cover completeness knobs in the check version contract #411 adds tests to this same file, so whichever lands second needs a quick rebase. Nothing in your rows overlaps.

Optional: the response_format row matches on a prefix only. Pinning the full message would catch a future rewording.

Good first PR, hope to see more.

[kstonekuan]

Merging with a fixup. Welcome, and the table itself is exactly right: every one of the nine refusals is now the only failure when its own condition is neutered, including both isinstance(..., bool) clauses separately from the type checks around them.

model empty / padded                       caught
response_format wrong type                 caught
api_key env var name malformed             caught
max_tokens not an int / <= 0               caught
max_retries not an int / negative          caught
temperature not finite                     caught
max_tokens bool guard alone                caught
max_retries bool guard alone               caught

uncovered refusals: 0

Those last two are the ones I most wanted and the easiest to leave out, because a test with max_tokens="32" looks like it covers the same ground and does not.

The fixup, dd5ea87. The file did not collect as submitted:

ERROR collecting tests/test_build_ai_vlm_checks.py
In test_openai_compatible_execution_accepts_valid_configuration: function uses no argument 'endpoint'

The twelve-case decorator had ended up on ..._accepts_valid_configuration, which takes no arguments, while the refusal test below still carried the old three-parameter signature and no decorator. Everything you wrote was right, it was assembled in the wrong order. I moved the decorator onto the refusal test, widened its signature to the eight fields the table supplies, and deleted the orphaned copy.

Worth saying plainly, because your PR description lists uv run pytest -q as passing: it does not pass on the submitted commit, it errors during collection, so the whole file is skipped rather than failing loudly. That is a nasty failure mode to hit as a newcomer, since a collection error scrolls past differently from a test failure. Nothing about it reflects on the work.

One thing to know rather than fix: #414 landed a couple of hours before this and covers max_retries=True too, so that one case is now asserted twice, and there are two positive-control tests. Harmless duplication, and I would rather leave both than delete someone else's test to tidy mine.

Gate on the merged result: ruff check, ruff format --check, ty check clean, 1557 passed / 6 skipped.

Closes #407. If you want another, the starter list is at https://github.com/Hebbian-Robotics/hflow/issues?q=is%3Aissue+is%3Aopen+label%3A%22good+first+issue%22, and #408 is the closest in shape to what you just did. Two tips for next time: only an assignee reserves an issue, so anything unassigned is fair game, and we cap non-collaborators at one open PR at a time so a bot will politely close a second and invite you back.