aislop v0.4.0 — hand off to your agent, fix what matters, skip what doesn't.

Hand off to any agent in one flag

When aislop fix resolves what it can automatically, you're often left with issues that need human (or agent) judgment — complex refactors, architectural decisions, security fixes that need context.

v0.4.0 generates a structured prompt with every remaining issue, affected files, and suggested fixes, then opens your preferred coding agent:

14 agents are supported out of the box: Claude Code, Codex CLI, Cursor, Windsurf, Gemini CLI, Amp, VS Code Copilot, Aider, Goose, OpenCode, Warp, Kimi Code, Antigravity, and Deep Agents. If your agent isn't listed, --prompt prints the context so you can paste it anywhere.

The prompt includes the scan score, every diagnostic with file path and line number, and a prioritized fix list. Agents get everything they need to fix issues without asking follow-up questions.

Fix that doesn't touch clean code

In previous versions, aislop fix ran every fixer unconditionally. Even if your code was clean, Biome would reformat files — producing noisy diffs like "40 insertions, 40 deletions" for what was just re-indentation.

v0.4.0 scans first, then only runs fixers for engines that found issues. If formatting reports 0 issues, the formatter doesn't run. If linting is clean, the lint fixer is skipped. This means:

• + No phantom diffs — clean code stays clean after fix
• + Faster runs — skips unnecessary tool invocations
• + Trustworthy output — "0 resolved" means 0 files changed, always

The output is cleaner too. When everything passes, the result box shows your score and stops. No more "Next steps: 1. All clear!" — if it's clear, we just say nothing.

Formatting issues are warnings, not errors

A tab being two spaces instead of four shouldn't block your PR. But that's exactly what happened — Biome reports formatting issues as error severity, and aislop was passing that through. Your score dropped from 100 to 94 because of indentation.

v0.4.0 downgrades all formatting issues to warning. This matches every other format engine in aislop (gofmt, ruff, rubocop, cargo fmt) which already used warning severity. The principle: auto-fixable style issues are warnings, real bugs are errors.

Your scores will be more accurate. A project with perfect logic but messy tabs gets 100, not 94.

Everything else

• + Diagnostic deduplication — no more duplicate issue reports from overlapping engines
• + Unused file removal in aislop fix
• + Directory existence validation in scan
• ~ Respect biome.json lineWidth config when scanning and fixing
• ~ Fallback to npm audit when pnpm audit endpoint is retired
• ~ Graceful config loading errors
• ~ CI upgraded to Node 22/24, pnpm/action-setup v5

Upgrade

Or if you have it installed:

Full changelog on GitHub.