deslop

deslop is a free Skill that helps coding agents audit and remove complexity accumulated through repeated implementation and correction cycles. It traces each candidate back to current requirements, real callers, public contracts, and independent behavioral evidence.

Its three priorities are redundant or self-referential tests, verification machinery whose producer and verifier share the same failure domain, and defensive or fallback layers without a current contract. The aim is a smaller codebase with meaningful behavior preserved.

Start with a read-only audit. Findings include evidence, confidence, and the behavior that must remain. Explicit apply mode permits edits within the requested scope; uncertain candidates stay unchanged. Security, persistence, protocols, supported compatibility, concurrency, resource limits, and scientific invariants are preserved unless direct evidence supports a change.

Usage: invoke $deslop audit in Codex or /deslop audit in a standalone Claude Code installation. Add paths to limit scope. Use tests audit to focus on test suites, or deep audit for a repository-wide review. After reviewing the findings, request apply for the chosen scope. The separate Claude Code Plugin uses /deslop:deslop.

This free package contains the unchanged v0.3.3 standalone Skill, its reference guides, and the MIT license. It requires a compatible coding agent and access to the target repository. Agent or model usage may incur charges from your provider.

Source, installation instructions, and version-bound evaluation records: https://github.com/MrZoyo/deslop-GPT. Published evaluations are development evidence; no general performance guarantee is claimed.