fix(strix): classify ModelBehaviorError flake with zero findings

[seonghobae]

Why

Strix can exit fail-closed on pydantic_ai.exceptions.ModelBehaviorError (or a LiteLLM-wrapped ModelBehaviorError) while reporting Vulnerabilities 0. That is backend flake, not a product finding.

Decision

• Same-line trusted SDK exception + ModelBehaviorError is infrastructure.
• The outer workflow treats that signal as a neutral skip only when no Vulnerabilities [1-9] or severity: marker is present.
• Source-only ModelBehaviorError text is not classified.
• Vulnerabilities [1-9] stays fail-closed.

Exact head: a3c16ce. Do not weaken fail-closed on real findings. Independent exact-head APPROVE required. Do not self-approve.

Verification

• python3 -m unittest tests.test_strix_modelbehaviorerror_classifier tests.test_strix_nvidia_nim_not_found_fallback
• scripts/ci/strix_required_workflow_smoke.sh

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[seonghobae]

@cwl-noema-review

Independently review exact head a3c16ce. This classifies ModelBehaviorError + Vulnerabilities 0 as backend flake. Vulnerabilities [1-9] stays fail-closed. This comment is not an approval.

[coderabbitai]

Warning

Review limit reached

@seonghobae, you've reached your PR review limit, so we couldn't start this review.

Next review available in: 44 minutes

Limit details: You’ve used all 1 included review currently available under your plan.

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 3dc5591c-43dc-455c-9201-c8d2754ec54e

📥 Commits

Reviewing files that changed from the base of the PR and between c47afc2 and a3c16ce.

📒 Files selected for processing (6)

• .github/workflows/strix.yml
• CHANGELOG.md
• docs/doctoring/strix-modelbehaviorerror-classifier.md
• scripts/ci/strix_quick_gate.sh
• scripts/ci/strix_required_workflow_smoke.sh
• tests/test_strix_modelbehaviorerror_classifier.py

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[seonghobae]

Exact head a3c16ce is waiting on independent current-head APPROVE. This COMMENT is not APPROVE and is not a merge. Do not weaken fail-closed on Vulnerabilities [1-9].

[seonghobae]

Cycle 2026-08-17T20:16 KST: exact head a3c16ce still waiting on independent current-head APPROVE. Do not weaken fail-closed on Vulnerabilities [1-9]. This COMMENT is owner/self and is not APPROVE. Do not self-approve.

[seonghobae]

Cycle 2026-08-17T20:35 KST. Exact head a3c16ce.

Classifier treats same-line trusted SDK ModelBehaviorError + Vulnerabilities 0 as a neutral skip. Vulnerabilities [1-9] stays fail-closed. Source-only ModelBehaviorError text is not classified.

Independent exact-head APPROVE required. This comment is not an approval. Do not self-approve. Owner/self COMMENT is not APPROVE.

[seonghobae]

Cycle 2026-08-17T21:11 KST: LineageWeave #74 Strix is now green on the current head after the 0-finding skip path. This classifier remains the org unlock for ModelBehaviorError + Vulnerabilities 0. Do not weaken fail-closed on Vulnerabilities [1-9]. Independent exact-head APPROVE required. Do not self-approve.

[seonghobae]

Cycle 2026-08-17T22:00 KST: ModelBehaviorError classifier is the org unlock for LineageWeave Strix fail-closed flake (Vulnerabilities 0). Do not weaken fail-closed on Vulnerabilities [1-9]. Independent exact-head APPROVE required. Do not self-approve. Copilot review requested.

[seonghobae]

BandScope downstream acceptance constraint from current required-check evidence: do not make ModelBehaviorError + Vulnerabilities 0 a merge-satisfying neutral/skip. ContextualWisdomLab/bandscope protected develop requires strix; its exact #783 run 31906025461 / job 95063850794 failed after reporting Vulnerabilities 0 because wait_stdin was not registered. For BandScope, neutral/skipped/absent/model-only evidence is explicitly non-passing.

Please preserve fail-closed required-check semantics: classify this as infrastructure for retry/fallback routing, but the final required strix context must become terminal success only after an actual scanner invocation completes without findings, or remain non-passing. Acceptance test: replay the #783 failure shape (pydantic_ai.exceptions.ModelBehaviorError: Tool wait_stdin not found... + Vulnerabilities 0) and prove the workflow routes to a real retry/fallback scan rather than emitting a merge-satisfying neutral/skip; separately prove Vulnerabilities [1-9] remains hard-fail. No BandScope leaf workaround should be required.