Professional Skills

Agent skills for professional writing, built for Claude Code, Codex, and skills.sh

Two production-grade skills that make anything you ship sound like a person wrote it.

Claude Code Plugin Codex Plugin Agent Skills License: MIT

npx skills add Kaos599/professional-skills --all

What's inside

Two skills, each a SKILL.md plus a references/ folder. They fail differently and run separately; the only coupling is that the writer calls the anti-slop skill as a hard gate before anything ships.

anti-slop-writing

Make AI text sound human. Rewrites drafts that read as machine-generated, restores the voice that AI editing flattened, and audits text for AI tells when asked.

Three modes:

  1. Improve (default): rewrite so it is better and sounds like a person.
  2. Audit: name the AI patterns, quote the line, give the fix. No rewrite, no score.
  3. Restore: rescue the person from a draft an AI already "polished" flat.

Why it works:

  • Structure outlives vocabulary. Word lists go stale; sentence-shape and paragraph-shape tells do not.
  • Tells count in clusters, never alone. One em dash means nothing; several pattern families co-occurring is a finding.
  • No AI-vs-human probability verdicts. Detectors guess, and they are most wrong against non-native English writers.
  • Subtraction is only half the job. A dedicated step puts the person back, because a stripped draft is bleached, not finished.
  • Ships with a read-only scanner (scripts/slop_scan.py) so the mechanical checks run in CI, not just in a model's head.

Use when: a draft feels generic or corporate, when polishing anything an LLM helped write, or before publishing anything.

technical-content-writer

Technical content that sounds like a specific human. Writes LinkedIn posts, blog posts, threads, newsletters, and essays for readers who can tell if you are faking it.

The pipeline:

0. Anchor     -> the concrete thing, the claim, the proof, the lesson
1. Calibrate  -> voice signature derived from exemplars at runtime
2. Shape      -> format mechanics and structure archetypes
3. Draft      -> in the signature, against the anchor
4. Sweep      -> five editing passes, each scored 0 to 10
5. Gate       -> anti-slop audit, hard threshold, nothing ships above the clean band
6. Ship

Why it works:

  • Voice is derived from samples at runtime. Drop a file into references/exemplars/ and the signature changes; no code edit.
  • Mechanism beats outcome. "Decode is bottlenecked on memory bandwidth because every step re-reads the weights" beats "decode is slow."
  • Every claim carries a number or a named thing. No anchor means no piece, and inventing a number, benchmark, or quote is absolute.

Use when: writing or rewriting any technical content for a public audience.

Install

skills.sh: any agent, one command

Works with Claude Code, Codex, and 30+ agents. Skills land in the right directory for your agent automatically.

# All skills, all agents
npx skills add Kaos599/professional-skills --all

# A single skill
npx skills add Kaos599/professional-skills --skill anti-slop-writing

# Globally, across all projects
npx skills add -g Kaos599/professional-skills --all

Claude Code: as a plugin

This repository is a valid Claude Code plugin with both skills bundled under skills/.

claude plugin marketplace add Kaos599/professional-skills
claude plugin install professional-skills

Or install the skills directly:

npx skills add Kaos599/professional-skills -a claude-code --all

Codex: as a plugin

This repository is a valid Codex / ChatGPT plugin (.codex-plugin/plugin.json) with a repo marketplace.

codex plugin marketplace add Kaos599/professional-skills

Then open Codex and install Professional Skills from the /plugins browser, or install the skills directly:

npx skills add Kaos599/professional-skills -a codex --all

[!TIP] Private repository? All three install paths work with your existing gh / git credentials; no extra setup needed.

Without installing

Point your agent at the folder, or copy the skill folders anywhere on the standard SKILL.md discovery paths (.claude/skills/, .agents/skills/, ...):

cp -R skills/anti-slop-writing ~/.agents/skills/
cp -R skills/technical-content-writer ~/.agents/skills/

Try it

Prompts that trigger the skills:

This draft reads like AI slop. Make it sound like me.            -> anti-slop-writing
Does this sound like it was written by a person? Be honest.      -> anti-slop-writing (audit)
Write a LinkedIn post about the inference stack we just shipped. -> technical-content-writer
Turn my README into a blog post that engineers will trust.       -> technical-content-writer

Repository structure

professional-skills/
├── .claude-plugin/          Claude Code plugin (plugin.json + marketplace.json)
├── .codex-plugin/           Codex / ChatGPT plugin manifest
├── .agents/plugins/         Codex repo marketplace
├── skills/
│   ├── anti-slop-writing/
│   │   ├── SKILL.md
│   │   ├── references/      banned vocabulary, structural patterns, rubric, code slop, worked examples
│   │   └── scripts/         slop_scan.py (read-only mechanical scanner)
│   └── technical-content-writer/
│       ├── SKILL.md
│       └── references/      voice protocol, playbooks, editing passes, exemplars
├── README.md
└── LICENSE

Both skills follow the standard SKILL.md + references/ Agent Skills layout, so any host (Claude Code, Codex, skills.sh, or any agent that reads SKILL.md files) picks them up automatically.

Adding a voice to technical-content-writer

Drop NN-short-slug.md into references/exemplars/ with the frontmatter header documented in that folder's README. Paste the piece verbatim, typos and all; those are signal. Set authored: true for your own writing, authored: false for writing you admire; the protocol weights them differently.

Provenance

The anti-slop and writing-craft content is distilled from a harvest of 100 published skills on skills.sh (~387,000 words across 45 anti-slop and 55 writing-craft skills), analyzed with ban-context document frequency so skills that ban a word are counted separately from skills that use it. The exemplars and house-style measurements come from a separate pass over real published technical writing, with sources cited per entry.

Professional Skills · MIT License · Built for Claude Code, Codex, and skills.sh

Agent Skills · Report an issue