arxiv-agentic-verifier
oleh wanng-ide
Coding Agents & IDEsRingkasan
**Source Paper:** [Scaling Agentic Verifier for Competitive Coding](https://arxiv.org/abs/2602.09012) (ID.
Deskripsi
Apa yang Dilakukan Skill Ini
An agentic verifier that checks code correctness by generating targeted test cases designed to expose logic flaws. It analyzes problem constraints and code structure to produce discriminative inputs rather than sampling randomly. Supports Python and JavaScript.
Generates discriminative test cases from problem analysis rather than random inputs, catching logic flaws that fuzzing and random sampling routinely miss.
Kapan Menggunakannya
- Verifying competitive programming submissions before upload
- Finding edge cases in coding interview solutions
- Catching off-by-one errors in algorithmic implementations
- Checking boundary behavior in math or sorting functions
- Automated validation of student code against problem specs
Contoh Alur Kerja
User asks: verify this Python solution for a two-integer sum problem
- 1
Parse the problem description to extract constraints and expected behavior
- 2
Analyze the submitted code's logic and input-handling structure
- 3
Generate targeted test cases aimed at boundary conditions and likely failure modes
- 4
Execute the code against each generated test in a sandboxed environment
- 5
Collect and report pass/fail results with the specific inputs that caused failures
Verification report listing which tests passed, which failed, and the exact inputs that broke the code
Persyaratan
- OPENAI_API_KEY environment variable
Instalasi
clawhub install wanng-ide/arxiv-agentic-verifierJalankan perintah CLI-nya, atau tempel prompt ke Claude Code, Cursor, Codex, atau coding agent apa pun yang Anda pakai.