What problem does it solve? AI-generated documentation, comments, and commit messages often contain inflated vocabulary, filler phrases, tautological docstrings, and robotic tone that make technical writing read as machine-produced. This Skill applies targeted fixes from a prior do-review-ai-writing run to make the text sound human. ## Core Features & Use Cases - Safe/Needs-Review Classification: Automatically partitions findings into auto-appliable safe fixes (deleting filler, swapping AI vocabulary, normalizing quotes) and interactive fixes requiring confirmation (rewriting promotional language, restructuring sections). - Hard Safety Gates: Verifies a clean git working tree, validates the review JSON shape and freshness against the current HEAD, and validates every modified file (markdown fence balance, Python AST parse, JS syntax check) with automatic revert on failure. - Dry-Run and Category Filtering: Preview all fixes with --dry-run, scan the full codebase with --all, or restrict to one category such as vocabulary, filler, or code_docs. - Use Case: After running do-review-ai-writing on a repository, invoke do-humanize to delete synthetic openers, swap words like "utilize" for "use", strip chat leaks from comments, and interactively rewrite promotional README language. ## Quick Start Ask the AI to run do-humanize with --dry-run to preview the fixes it would apply from the existing ai-writing-review.json findings.
