test(guardrails): stop the onyx and prompt security tests leaking env vars on failure

[yuneng-berri]

TLDR

Problem this solves:

• Five guardrail test files write straight to os.environ
• None has a try/finally, so a failure leaves them set
• Two of the config tests asserted nothing at all

How it solves it:

• monkeypatch.setenv and delenv unset on failure too
• One autouse fixture replaces setup and teardown methods
• Those two config tests now pin what registration produces

User Flow

No end-user behavior changes. A proxy admin running any of these five guardrails
gets the same request checked the same way; these are the test files around
those guardrails, not the guardrails

Relevant issues

Linear ticket

Pre-Submission checklist

• I have added meaningful tests
• The handful of test files covering my change pass locally
• My PR passes all required CI/CD checks (e.g., lint, schema.d.ts sync check, etc.)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review

Screenshots / Proof of Fix

Shared setup: clear the keys, run one test with one line forced to fail, read
the environment back in the same interpreter. A passing run is clean on both
sides, so the forced failure is the only state where they differ. probe is
this, parameterised by test id and keys:

probe() { uv run python -c '
import os, pytest, sys
keys = sys.argv[2:]
for k in keys: os.environ.pop(k, None)
pytest.main([sys.argv[1], "-q", "-p", "no:randomly"])
print("env after the run:", {k: os.environ[k] for k in keys if k in os.environ} or "clean")
' "$@"; }

$FILES below is the five test files this PR touches.

Before (ff02d5c)

1. onyx

1. Add assert False, "forced" under assert timeout_param.read == 45.0, then
   probe tests/test_litellm/proxy/guardrails/guardrail_hooks/test_onyx.py::test_onyx_guard_with_custom_timeout_from_kwargs ONYX_API_BASE ONYX_API_KEY ONYX_TIMEOUT

1 failed, 2 warnings in 1.87s
env after the run: {'ONYX_API_BASE': 'https://test.onyx.security', 'ONYX_API_KEY': 'test-api-key'}

2. hiddenlayer

1. Add raise AssertionError("forced") above the init_guardrails_v2( call in
   test_hiddenlayer_config_saas, then
   probe tests/test_litellm/proxy/guardrails/guardrail_hooks/test_hiddenlayer.py::test_hiddenlayer_config_saas HIDDENLAYER_API_BASE HIDDENLAYER_API_KEY

env after the run: {'HIDDENLAYER_API_BASE': 'https://my.hiddenlayer'}

3. prompt security

1. Add assert False, "forced" under the first assert of
   test_apply_guardrail_block_request, then
   probe tests/test_litellm/proxy/guardrails/test_prompt_security_guardrails.py::test_apply_guardrail_block_request PROMPT_SECURITY_API_KEY PROMPT_SECURITY_API_BASE

1 failed, 2 warnings in 1.87s
env after the run: {'PROMPT_SECURITY_API_KEY': 'test-key', 'PROMPT_SECURITY_API_BASE': 'https://test.prompt.security'}

4. what the quality rules see

1. for f in $FILES; do uv run python scripts/check_test_quality.py $f; done | grep -oE "TQ00[0-9]" | sort | uniq -c

   7 TQ001
   1 TQ002
   5 TQ003
  89 TQ004
  13 TQ005

After (15138fd)

1. onyx

1. Same injection, same probe

1 failed, 2 warnings in 1.83s
env after the run: clean

2. hiddenlayer

1. Same injection, same probe

env after the run: clean

3. prompt security

1. Same injection, same probe

1 failed, 2 warnings in 1.86s
env after the run: clean

4. what the quality rules see

1. for f in $FILES; do uv run python scripts/check_test_quality.py $f; done | grep -oE "TQ00[0-9]" | sort | uniq -c

   5 TQ001
   1 TQ002

2. uv run pytest tests/test_litellm/proxy/guardrails/ -q -p no:randomly -n 4

2873 passed, 1 xfailed, 20 warnings in 14.15s

Type

🐛 Bug Fix

✅ Test

Caveats (if any)

• The leak only shows on failure, which is when it matters
• Nothing under proxy/guardrails relied on it, all 2873 pass
• Onyx and prompt security config tests gained real assertions
• The other three keep their zero-assert tests, which need judgement
• 42 more files outside proxy/guardrails have the same shape
• Touches test-quality-budget.json, as do test: drop the cwd-relative sys.path.insert calls from the test suite #37802, test: unwind environment writes in tests/test_litellm with monkeypatch #37806, test(cost-calc): stop 182 global writes leaking out of the cost-calc suites #37815, test(mcp): retire the last file of the dead tests/litellm mirror #37824 and test(callbacks): unwind the callbacks global the policy engine and realtime tests scaffold around #37826

Final Attestation

• The tests check the right things, including the edge cases, and regressions in the respective real-world customer use-cases are not possible after this PR

[greptile-apps]

Greptile Summary

This PR improves guardrail-test isolation by replacing direct environment and global-state mutations with pytest monkeypatch cleanup.

• Prevents environment variables from leaking when guardrail tests fail.
• Adds registration assertions to the Onyx and Prompt Security configuration tests.
• Tightens the test-quality budget to reflect the removed violations.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
test-quality-budget.json	Lowers test-quality limits to account for violations removed by the test cleanup.
tests/test_litellm/proxy/guardrails/guardrail_hooks/test_deepkeep.py	Uses monkeypatch for environment and global-state isolation.
tests/test_litellm/proxy/guardrails/guardrail_hooks/test_hiddenlayer.py	Replaces direct environment mutations with failure-safe monkeypatch operations.
tests/test_litellm/proxy/guardrails/guardrail_hooks/test_onyx.py	Adds failure-safe environment isolation and verifies the configured guardrail registration; the previously reported line-length issue is resolved.
tests/test_litellm/proxy/guardrails/guardrail_hooks/test_repelloai.py	Isolates environment and global-state changes through monkeypatch.
tests/test_litellm/proxy/guardrails/test_prompt_security_guardrails.py	Makes environment cleanup failure-safe and adds concrete registration assertions.

_{Reviews (3): Last reviewed commit: "test(guardrails): stop five guardrail te..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[yuneng-berri]

@greptileai

[yuneng-berri]

@greptileai