[vanshwadhwa24]

Problem and scope

guardrails_runner.py check crashes with an uncaught RecursionError (traceback, exit code 1) on deeply nested configuration JSON, breaking the documented contract: exit 2 for file/schema parsing failures, always a JSON report on stdout (reported in #400). A CI gate cannot distinguish a crashed parser from a failed validation. Scope is the runner's error handling plus regression tests.

What changed

• guardrails_runner.py: main() now catches RecursionError alongside the existing OSError, ValueError, TypeError, so a deeply nested config produces the standard JSON error report on stdout and exit code 2. In my reproduction the RecursionError is raised by json.loads during parsing; the same handler would also catch one raised later in the command path, though I have only reproduced the parsing case.
• Added test_guardrails_runner.py next to the runner with two regression tests: a 10,000-level config (60,001 bytes, under the 1 MB limit) must exit 2 with a parseable JSON report and no traceback on stderr, and the shipped guardrails_chatbot.json profile must still pass with exit 0.

Verification

• python -m black --config .github/quality/pyproject.toml --check .github/quality/tests
• python -m flake8 --config .github/quality/.flake8 .github/quality/tests
• python -m pytest -c .github/quality/pyproject.toml .github/quality/tests (924 passed; this suite does not discover the new test file)
• I used UnvibeCode on a representative repository when the change affects code analysis, CLI behaviour, engineering guidance, or code-specific claims. (Left unchecked deliberately: the change is a one-line error-handling fix in a shipped example script, and the code-specific claims in this PR are verified by the direct test runs below rather than by an UnvibeCode review, which would not exercise this behavior.)
• I updated documentation for changed user-facing behaviour. (Not applicable: this restores the already-documented exit-2 contract.)
• I included no credentials, proprietary source, customer data, generated customer reports, or unsanitized logs.

Changed-file checks: the new test_guardrails_runner.py passes black --check and flake8 with the repo configs. The modified guardrails_runner.py is a one-line change to an existing statement, left in the file's current style; note CI currently lints only .github/quality/tests, so neither skills/ file is covered by CI lint (#402 tracks that gap).

Direct test runs (from skills/unvibecode_gaurdrails_implemenataion_pack/Pre-configured_gaurdrails_programs/, python -m pytest test_guardrails_runner.py -q):

• On unpatched main: 1 failed, 1 passed - the deeply nested config (60,001 bytes, 10,000 levels) crashes with RecursionError: maximum recursion depth exceeded while decoding a JSON object, traceback on stderr, exit code 1, no stdout report.
• On this branch: 2 passed - the same config prints {"configuration_valid": false, "runtime_verified": false, "error": "Invalid configuration or file operation; check the schema and path."} on stdout and exits 2.

Disclosure: this change was prepared with AI assistance (code edit and test runs); the before/after results above are the actual outputs from the local runs.

Fixes #400