I built an open-source, offline AST guardrail & MCP server (Aveyx v3.0.1) to stop AI coding agents from faking tests, leaking secrets, and breaking prod
If you use Claude Code, Cursor, Windsurf, or Copilot on a multi-file repo, you've probably seen them: Swallow failing assertions or slip in it.skip / || exit 0 to make tests "pass" Edit a shared helper without checking
If you use Claude Code, Cursor, Windsurf, or Copilot on a multi-file repo, you've probably seen them:
- Swallow failing assertions or slip in
it.skip/|| exit 0to make tests "pass" - Edit a shared helper without checking what breaks downstream
- Hardcode an API key or leave raw SQL/command concatenation behind
- Burn context window dumping full implementation files
I just shipped Aveyx v3.0.1 (MIT, Node 20+ ESM, 100% offline, only 3 @babel/* runtime deps) to catch these mechanically instead of relying on prompt rules alone:
-
aveyx impact <file>— Computes transitive reverse dependencies (ESM, CJS, re-exports, dynamicimport()) and lists every affected test file. -
aveyx compact— Strips function bodies via AST while preserving signatures, types, and PageRank callers. -
aveyx review --strict— Babel AST anti-tamper + secret scanner + protected-config guard (outputs OASIS SARIF 2.1.0). -
aveyx audit-bugs&aveyx scan-secrets— AST SQLi / command injection / React hook leak + credential scanner. -
aveyx checkpoint/rollback— Atomic snapshots (including.envand git hooks) before risky refactors. -
aveyx install-hooks --pre-push— Blocks bad commits/pushes locally. -
aveyx-mcp— Native stdio MCP server with 10 tools and 44 prompts.
Quick start:
bash
npx aveyx init
npx aveyx doctor

📰 Read the original article on Dev.to AI
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.