Quantum code review
AI writes quantum code. qcheck reviews it.
A review layer for AI-generated Qiskit and OpenQASM.
Catch the avoidable before code reaches humans, CI, or simulators.
Signals
What qcheck reviews
qcheck v0 reviews static signals — no execution required, so it runs safely on untrusted model output inside an agent loop or CI.
API usage
Flags removed-in-1.0 imports like execute() and Aer, and deprecated gate aliases.
Parse issues
Header, register declarations, index ranges, and measurement validity.
Unsafe patterns
Filesystem, network, process, or dynamic-exec constructs are flagged before anything runs.
Missing measurements
Common mistakes that make a circuit fail or return nothing useful.
Common LLM mistakes
The recurring errors models make in generated quantum snippets.
Review
A review, not a verdict
Each snippet returns a set of signals — passed or flagged — that a developer or agent can act on before the code runs.
Scope
What it does — and doesn't
In scope
- Static review of Qiskit / OpenQASM snippets
- Common, avoidable failures
- CI and agent preflight
- Machine-readable JSON output
Not in scope
- Hardware execution
- Algorithm correctness or formal proof
- PQC migration
Benchmark
Static review benchmark
Tracks how AI-generated snippets perform against qcheck's current static review checks. It reports static_pass_rate on a small public task set as an early quality signal. Rows marked SAMPLE are demo data.
| # | Model | Provider | Static pass rate | Passed / Attempted | Unsafe | qcheck |
|---|---|---|---|---|---|---|
| 1 | demo-model SAMPLE | sample | 50.0% | 3 / 6 | 1 | v0.1.0 |
Real runs will include provenance, task counts, qcheck version, and prompt hash. See the methodology.
Roadmap
Where it is going
CLI
Static review for Qiskit and OpenQASM, with JSON output for agents and CI.
GitHub Action & MCP
Review in CI and as an agent-callable tool.
Hosted API
A review endpoint for agent workflows.