skill
mega-anti-slop
Evidence-led Agent Skill for removing generic AI patterns from email, product copy, social writing, and long-form prose.
Set up
Skills are installed by placing their folder where your agent loads skills.
The download comes from github.com/sanketagarwal/mega-anti-slop itself, not from our server, so what you get is exactly what its author published.
Install instructions come from the README
README
synced Oct 23, 2026Mega Anti-Slop
An Agent Skill for making professional writing more specific, natural, and appropriate to its context—without forcing everything into one “humanized” voice.
It works with emails, web copy, product UI, social posts, blogs, newsletters, announcements, and bios.
Install
npx skills add sanketagarwal/mega-anti-slop
That is the only installation command. The Agent Skills CLI detects your compatible agent and installs the skill in the right place.
Use
Use $mega-anti-slop to revise this draft. Preserve the facts, uncertainty, and my voice. Return only the finished copy.
What it does
- Protects facts. Records names, dates, quantities, limitations, uncertainty, and requested actions before editing.
- Routes by context. Uses separate guidance for email, web copy, publishing, and product UI.
- Repairs the real problem. Finds the exact weak span and replaces vague claims with supplied facts, mechanisms, consequences, or concrete actions.
- Checks the result. Verifies factual fidelity, tone, clarity, rhythm, formatting, and visible limitations.
Patterns are diagnostic signals, not automatic bans. A long sentence, fragment, passive construction, group of three, or em dash can stay when it is accurate and useful.
Before and after
Before:
TraceLock is a cutting-edge platform that empowers modern security teams to streamline SOC 2 compliance with seamless integrations.
After, using supplied product facts:
TraceLock uses read-only OAuth scopes to collect repository settings, tickets, and access events from GitHub, Jira, and Okta, then maps them to a review queue. Your team approves each item before export.
The rewrite explains the systems, access model, product behavior, and human control. It does not merely swap one set of fashionable words for another. See the complete fact audit.
Modes
| Mode | Examples | Protects |
|---|---|---|
| Follow-ups, outreach, scheduling | Relationship, courtesy, concrete requests | |
| Web copy | Landing, feature, and pricing pages | Audience, mechanism, proof, limitations |
| Publishing | Social, blogs, newsletters, bios | Channel fit, bounded claims, nuance |
| Product UI | Labels, errors, empty states | Clarity, recovery, terminology |
What it will not do
- Guess whether AI or a person wrote something
- Evade AI detectors or prove human authorship
- Invent facts, metrics, quotes, urgency, stories, or personality
- Ban punctuation, long sentences, technical terms, or courtesy
- Apply one house style to every writer and channel
Evidence
- Five original worked examples
- Independent forward-test report
- Reproducible benchmark protocol
- Exact benchmark prompts
- Research and license review
- Source credits
The benchmark protocol is ready for blinded human review. This repository does not claim a benchmark win before those runs are complete.
Repository map
| Path | Purpose |
|---|---|
SKILL.md | Core workflow and routing |
references/ | Artifact guides and pattern diagnostics |
examples/ | Rewrites with line-by-line fact audits |
docs/ | Research, validation, and benchmark material |
scripts/validate.py | Dependency-free repository validator |
Only SKILL.md and the relevant artifact reference enter the agent context. Repository documentation stays outside the prompt.
Limitations
- English-first; multilingual behavior is not validated.
- It preserves supplied facts but cannot verify them or replace missing research.
- Results vary by model and agent harness.
- Legal, academic, crisis, fiction, and localization work need specialist guidance.
- Voice preservation is limited without a representative writing sample.
Contributing and license
See CONTRIBUTING.md for contribution requirements and SECURITY.md for private vulnerability reporting.
MIT © 2026 Sanket Agarwal.