What problem does it solve?
Maintainers of the NVIDIA/NemoClaw repository accumulate stale bug reports whose status on current releases is unknown. This Skill automates the verification loop: it selects old Bug-type issues, reproduces them on the reported release, re-tests on the newest release tag, and prepares an evidence-backed comment and Project update for maintainer approval.
Core Features & Use Cases
- Candidate selection and filtering: Queries open issues via GitHub GraphQL, filters by native Issue Type, platform, security, and idempotency markers, and parses the reported version from labels or body text.
- Safe reproduction on Brev instances: Treats issue reproducers as untrusted input, builds a bounded reviewed reproducer, provisions or reuses Brev CPU/GPU instances with a 60-minute budget, and validates the reported release before testing the newest tag.
- Evidence-backed verdicts with approval gates: Scores confidence, redacts transcripts, drafts concise comments (fixed-on-latest, still-reproduces, by-design, verify-inconclusive), and never writes to GitHub or closes issues without explicit maintainer approval.
- Use Case: A maintainer asks to drain the bug backlog; the Skill verifies issue #2604 against v0.0.34, confirms the bug is fixed, and proposes a tag-anchored comment with linked evidence.
Quick Start
Ask the agent to verify whether stale NemoClaw bug issue number 2604 still reproduces on the newest release tag.