Launched this week
Parry
Catch bugs in AI-written code before you push
6 followers
Catch bugs in AI-written code before you push
6 followers
Traditional linters check syntax and human typos; they do not understand probabilistic AI code. Parry runs locally in < 1ms to catch silent signer downgrades, hallucinated package imports (slopsquatting), and prompt-to-transaction wallet drains.
Free Options
Launch Team
Build your Framer site from Claude, Codex or CursorYour AI agent builds it. You review before it goes live.
Promoted
Hey everyone, I'm Shahbaz, the creator of Parry.
A few months ago, I was refactoring a Solana program with Claude in Cursor. Claude hit an Anchor trait bound conflict and resolved it by changing pub authority: Signer<'info> to pub authority: AccountInfo<'info>.
cargo check passed with zero warnings. The syntax was valid. But that change removed signature verification, meaning anyone with the public key could have drained the vault.
Compilers like tsc, eslint, and cargo check only verify syntax and types. They cannot tell when an AI model downgrades an authorization check to make an error go away, drops error handling, or imports a hallucinated package that an attacker registered on npm.
I built Parry to catch these patterns before code leaves localhost. It runs as a local pre-commit check in under 1 millisecond on Python, TypeScript, JavaScript, and Rust. It uses deterministic pattern rules rather than calling an LLM API, so it adds zero lag to your workflow and works completely offline.
You can run the free CLI locally right now: curl -fsSL https://parry.scopeledger.net/install.sh | sh
I would appreciate your feedback on two specific questions:
What is the nastiest silent bug an AI tool has introduced into your repo?
What frameworks or security rules should we add next?
If you drop a link to your public repository in the comments below, I will run our triage analyzer across it and reply with the report.