A deterministic score for AI-generated mess

What it does

aislop is a deterministic static analyzer that scans code for 50-plus anti-patterns across eight languages, including TypeScript, Python, Go, and Rust. It looks for the specific mess AI agents tend to produce: self-explanatory code prefaced by narrative comments, hidden fallbacks, hallucinated imports, duplicated helpers, and oversized functions. Every change gets a 0–100 score, and the tool runs sub-second without an LLM in the runtime path.

The interesting bit

Rather than blindly trusting an LLM to judge code quality, aislop uses hard-coded rules to detect AI slop deterministically—same code in, same score out. If your repository is mostly an unsupported language like C or Swift, it deliberately withholds the score instead of misrepresenting a codebase it never read. It can also spawn a local git worktree and hand off repairs to a headless coding agent like Codex or Claude Code, then verify the result before you review the diff.

Key highlights

• 50+ rules across 8 language targets, including Expo and React Native variants for JavaScript/TypeScript
• Sub-second scans with deterministic scoring; no LLM API calls at runtime
• Outputs JSON, SARIF 2.1.0, and public README badges
• Auto-fixes mechanical issues (unused imports, dead code) and can create agent handoff prompts for trickier problems
• Per-rule severity overrides and inline suppressions via aislop-ignore-next-line comments

Caveats

• The agent-repair feature depends on having a supported coding agent CLI already installed locally
• Repositories dominated by unsupported languages (C, C++, C#, Swift, Kotlin, and others) receive no score rather than a misleading partial one

Verdict

Worth a look if your team is shipping AI-generated code and wants a fast, deterministic sanity check that won’t rack up API bills. Skip it if your stack is outside the eight supported languages or you already have a custom lint pipeline that catches narrative comments and hidden fallbacks.