Zero-Slop: A Skill to Kill AI-Sounding Prose

Where this fits

You ask Claude to sharpen a client update. It returns 'utilize' and 'leverage'—clean, polished, but somehow less you. Zero Slop is a skill that lets your existing assistant edit the draft in context, while local tools point to exact phrases and protect the source. It fits the loop of generating prose, spotting AI-sounding patterns, and fixing them without moving to a separate model.

Why agents benefit

• The assistant reads the full draft in context, so edits respect the original facts, tone, and format instead of applying a generic rewrite.
• Local tools return exact phrases and line references, so the agent can show the user what changed and why rather than saying 'I improved it.'
• The workflow separates inspection from rewriting: the agent can detect AI-sounding patterns first and leave the prose untouched if that is what the user wants.
• Repeated, reason-labelled human edits become private learning signals, so future edits align with the user's preferences without prompt engineering.
• Before/after measurements give a concrete quality check before publishing, making the review less subjective.

Practical scenarios

Final pass on an agent-generated blog post

An agent drafts a post from outline to export. Before publishing, the user asks for a humanization pass. Zero Slop flags repeated phrases, paratactic constructions, and 'delve'-style words; the agent rewrites in context while preserving the outline and facts.

LinkedIn and social copy that sounds like a person

A developer writes a career update, then runs it through the assistant. Zero Slop's inspection tools identify sentences where every clause is the same length and the 'AI accent' is strongest. The agent keeps the substance but tightens rhythm and vocabulary.

Private polish for client-facing notes

Consultants often cannot paste client details into random external tools. Because Zero Slop operates inside the user's existing assistant, the text stays in the same environment, and local tools only measure patterns. The agent edits in context without uploading to a separate service.

Add it to your agent workflow

Install with:

npx skills add manavmishra/ZeroSlop --skill zero-slop

Then use it like any model capability. For example, ask: 'Apply zero-slop to this draft: flag the AI-sounding phrases, rewrite them, and keep the original version intact.' The assistant reads the content, local tools point to exact phrases, and the user gets a change list, not a wholesale rewrite. The skill works with Claude, GPT, or another compatible assistant, so there is no separate model or account to manage.

Compare before adopting

Before adopting it permanently, compare three things. Quality: run a side-by-side test on a sample with deliberate AI-sounding phrases, then check whether the edits keep the user's voice. Freshness: look at repository activity and release version. Alternatives: a well-crafted prompt or a simple linter may be enough for a simple use case. Workflow fit: confirm your agent setup supports skills and local tool execution. The skill earns its place when the editing loop is frequent enough to justify the setup.

Why it is worth tracking

Zero Slop addresses a recurring problem: generated text that is factually correct but stylistically flat. With version 2.7.7, 20 stars, and an MIT license, it is early yet active enough to test. Watch the issue tracker and release notes for model-specific changes and growing local tool coverage. If your users ask 'make it sound less like ChatGPT' more than once a week, evaluate this skill now. The OpenAgentSkill entry is at https://www.openagentskill.com/skills/manavmishra-zero-slop.