SkillAgentSearch skills...

Deslopify

Make the AI writing sound more humane

Install / Use

/learn @shreyas-makes/Deslopify
About this skill

Quality Score

0/100

Supported Platforms

Universal

README

deslopify

Deslopify is an agent-neutral skill (Agent Skills format) that rewrites AI-sounding prose into natural human copy while preserving Markdown structure and factual content.

What it does

  • Removes formulaic AI writing patterns using a 24-pattern taxonomy.
  • Preserves protected spans: code blocks, inline code, URLs, link text, quotes, numbers, citations, and footnotes.
  • Uses a hybrid pattern source strategy:
    • curated local rules (canonical)
    • Wikipedia sync cache (update signal)
  • Defaults to final rewritten text only unless diagnostics are requested.
  • Ships a real CLI: deslopify lint, deslopify check, and deslopify eval.
  • Supports explicit rewrite presets: technical, marketing, essay, and team-update.
  • Excludes protected spans from lint and evaluation scans by default.

Skill layout

skills/deslopify/
├── __init__.py
├── cli.py
├── core.py
├── SKILL.md
├── references/
│   ├── ai_tells.wikitext
│   ├── ai_tells_fallback.txt
│   ├── pattern-catalog.md
│   ├── patterns.json
│   ├── rewrite-protocol.md
│   └── voice-mode.md
├── scripts/
│   ├── check_rewrite.py
│   ├── evaluate_rewrites.py
│   ├── fetch_wikipedia_ai_tells.py
│   └── lint_ai_tells.py
└── tests/
    └── fixtures/
        ├── input/
        └── output/

Install the CLI

python3 -m pip install .

For local development:

python3 -m pip install -e .

This installs the deslopify command.

Install the skill

Codex

mkdir -p ~/.codex/skills
cp -R skills/deslopify ~/.codex/skills/deslopify

Then restart Codex.

Claude / Copilot

Copy skills/deslopify into the configured skills directory for your setup, then restart the agent.

Use the skill

Prompt examples:

  • deslopify article.md
  • deslopify this draft and keep headings as-is
  • deslopify this memo with the technical preset
  • deslopify this team update and include diagnostics

Use the CLI

Lint a draft:

deslopify lint path/to/draft.md
deslopify lint path/to/draft.md --diagnostics --preset technical
deslopify lint path/to/draft.md --format json --strict

Validate a single rewrite pair:

deslopify check before.md after.md
deslopify check before.md after.md --preset team-update

Evaluate the fixture suite:

deslopify eval skills/deslopify/tests/fixtures
deslopify eval skills/deslopify/tests/fixtures --format json

Script commands

Run from repo root.

Refresh AI-tell cache

python3 skills/deslopify/scripts/fetch_wikipedia_ai_tells.py

Behavior:

  1. live fetch from Wikipedia
  2. fallback to cached skills/deslopify/references/ai_tells.wikitext
  3. fallback to local skills/deslopify/references/ai_tells_fallback.txt

Lint a draft

python3 skills/deslopify/scripts/lint_ai_tells.py path/to/draft.md

Useful options:

python3 skills/deslopify/scripts/lint_ai_tells.py path/to/draft.md --strict
python3 skills/deslopify/scripts/lint_ai_tells.py path/to/draft.md --format json
python3 skills/deslopify/scripts/lint_ai_tells.py path/to/draft.md --min-severity high
python3 skills/deslopify/scripts/lint_ai_tells.py path/to/draft.md --diagnostics --preset technical

Exit behavior:

  • 0: no blocking patterns
  • 2: blocking patterns found
  • 1: runtime/config error

Evaluate fixtures

python3 skills/deslopify/scripts/evaluate_rewrites.py skills/deslopify/tests/fixtures

Check one rewrite pair

python3 skills/deslopify/scripts/check_rewrite.py before.md after.md

Checks:

  • protected-span invariants unchanged
  • pattern-hit reduction from input to output
  • length drift threshold guardrail

Authoring rules enforced by the skill

  • Keep headings/lists/tables/order intact unless explicitly requested otherwise.
  • Never invent facts or citations.
  • Prefer direct, concrete phrasing over inflated or promotional language.
  • Keep tone aligned with source context (technical, marketing, essay).

References

View on GitHub
GitHub Stars4
CategoryContent
Updated10d ago
Forks1

Languages

Python

Security Score

70/100

Audited on Mar 29, 2026

No findings