Skip to content
← Back to all skills
Writingsafecommunity

unslop

Post-process AI-generated text through the unslop CLI to strip AI writing patterns before publishing

unslop — Strip AI Writing Patterns via CLI

Overview

unslop is a CLI tool that post-processes text to remove AI writing patterns programmatically. Unlike skills that ask the agent to avoid AI-isms, unslop runs as a deterministic pipeline step: pipe text in, get clean text out. Use it as a final pass before committing docs, publishing posts, or sending any AI-generated content to production.

The --deterministic flag makes output reproducible — same input always produces same output. The --stdin flag reads from stdin, enabling shell pipeline composition.

When to Use This Skill

  • When you have AI-generated text ready to publish and want a final cleanup pass
  • When working in a shell pipeline where text quality needs to be enforced automatically
  • When writing commit hooks or CI steps that validate content before it ships
  • When you need reproducible text normalization across multiple runs

Setup

Install once:

pipx install unslop
# or
uv tool install unslop

Verify:

unslop --version

How It Works

Step 1: Pipe Text Through unslop

Standard cleanup (may vary slightly between runs):

echo "This leverages cutting-edge AI to deliver robust solutions." | unslop --stdin

Deterministic cleanup (same input → same output every run):

echo "This leverages cutting-edge AI to deliver robust solutions." | unslop --stdin --deterministic

Step 2: Use in Shell Pipelines

Pipe the output of any command through unslop:

cat draft.md | unslop --stdin --deterministic > clean.md

Or chain with other tools:

cat draft.md | unslop --stdin --deterministic | pbcopy   # macOS: copy clean text to clipboard

Step 3: Integrate into Commit Hooks or CI

Add to a pre-commit hook or CI step to enforce quality gates on any generated content before it ships:

# In .git/hooks/pre-commit or a CI script
CONTENT=$(cat docs/changelog.md)
CLEANED=$(echo "$CONTENT" | unslop --stdin --deterministic)
if [ "$CONTENT" != "$CLEANED" ]; then
  echo "Changelog contains AI writing patterns. Run: cat docs/changelog.md | unslop --stdin --deterministic > docs/changelog.md"
  exit 1
fi

Examples

Example 1: Clean a Draft Document

cat blog-post-draft.md | unslop --stdin --deterministic > blog-post-final.md

Example 2: Inline Cleanup During Writing

# Write content, pipe through unslop, write result back
cat README.md | unslop --stdin > README.clean.md && mv README.clean.md README.md

Example 3: Validate Before Submitting a PR

# Check if any generated docs need cleanup
for f in docs/*.md; do
  ORIGINAL=$(cat "$f")
  CLEANED=$(echo "$ORIGINAL" | unslop --stdin --deterministic)
  [ "$ORIGINAL" != "$CLEANED" ] && echo "Needs cleanup: $f"
done

Best Practices

  • ✅ Use --deterministic in CI and automation to ensure reproducible output
  • ✅ Run on the final draft, not intermediate iterations
  • ✅ Combine with the avoid-ai-writing skill for both generation-time guidance and post-processing
  • ❌ Don't run on code files — unslop targets prose, not source code
  • ❌ Don't skip review after unslop: automated cleanup can occasionally change meaning; read the output

Limitations

  • Processes prose only — not code, JSON, or structured data
  • Does not catch factual errors or substantive writing issues
  • Some replacements may not fit every context; review the output before publishing
  • Requires Python tooling such as pipx or uv for standalone CLI installation

Security & Safety Notes

  • unslop reads from stdin and writes to stdout — no file system side effects by default
  • --deterministic mode is local and does not make LLM API calls
  • Default LLM mode may use ANTHROPIC_API_KEY or the Claude CLI; use --deterministic for sensitive local files and CI gates
  • Safe to run in CI pipelines and commit hooks when pinned to deterministic mode
— Field Manual

The 1,441 skills, demystified in one PDF.

A free editorial guide we wrote for the Skills Atlas: taxonomy, the 25 must-know skills, anti-patterns, and learning paths by profile.

  • 70+ pages, table of contents, ready to print.
  • Sent by email — link valid 7 days.
  • You can unsubscribe in one click anytime.

No spam. We never share your email. Unsubscribe in one click.