unsloppify

name	unsloppify
description	Strip AI writing patterns from prose. Run the unsloppify CLI to auto-fix the deterministic stuff (em dashes, banned phrases, smart quotes), then rewrite structural cliches by hand. Use when drafting, editing, or reviewing markdown that needs to not read as AI-generated.
metadata	{"trigger":"Editing markdown, blog drafts, READMEs, or any prose where the user says \"unsloppify\", \"remove AI tells\", \"anti-slop\", \"stop slop\", or \"make this sound human\""}

unsloppify

Strip characteristic AI writing patterns from prose. This skill pairs with the unsloppify Python CLI, which lives at https://github.com/petems/unsloppify and is installable via uvx unsloppify.

Workflow

When asked to unsloppify a piece of prose:

1. Hard-lock pass with the CLI. Run uvx unsloppify <file> --fix on the target file. This auto-rewrites unambiguous catches: em dashes become commas, "lean into" becomes "embrace", smart quotes become straight quotes, and so on. Show the user the diff.
2. Re-scan for findings the CLI can detect but not fix. Run uvx unsloppify <file> (no --fix). The remaining findings are judgment calls: banned vocabulary (delve, tapestry), sentence openers (Moreover,, Furthermore,), structural cliches (binary contrasts, false agency).
3. Judgment pass: rewrite each remaining finding. For each one, decide whether to delete, rephrase, or leave (some are genuinely the right word in context). Apply the rules in references/ to guide rewrites. Don't ask permission for each one — make the call and show the user the result.
4. Self-check with the rubric. Score the rewritten prose using references/rubric.md. If it scores below 35/50, revise again.
5. Verify clean. Final pass: uvx unsloppify <file> should exit 0 (or only have rule IDs you've consciously chosen to keep).

CLI quick reference

uvx unsloppify path.md                   # lint, exit 1 on findings
uvx unsloppify path.md --fix             # rewrite safe deterministic fixes
uvx unsloppify path.md --format json     # machine-readable
uvx unsloppify --list-rules              # show all bundled rules
uvx unsloppify path.md --severity error  # only errors

If uvx isn't available, fall back to pipx install unsloppify then unsloppify.

What gets auto-fixed (tier 1 + 2)

These are handled by the CLI's --fix mode — you don't have to think about them:

• Punctuation: em dashes (—), en dashes in prose, smart quotes, ellipsis chars, non-breaking spaces, zero-width characters, double exclamation marks
• Business jargon swaps: lean into → embrace, circle back → revisit, deep dive → examination, moving forward → next, utilize → use

What you have to judgment-call (tier 3)

These the CLI flags but won't fix — that's your job:

• Banned vocabulary (delve, tapestry, pivotal, vibrant, bolster, garner, underscore, multifaceted, foster, leverage, paramount, groundbreaking, cutting-edge, transformative, seamless, aforementioned, harness, spearhead, unprecedented, profound, empower). See references/phrases.md.
• Sentence openers (Moreover,, Furthermore,, Additionally,, Interestingly,, Notably,, Importantly,, Indeed,, Overall,, Certainly,, Absolutely,, Great question!, As an AI, As a language model, I'd be happy to).
• Throat-clearing openers (Here's the thing, Let me be clear, The truth is, It turns out, The uncomfortable truth is).
• Structural cliches — see references/structures.md:
  • Binary contrasts ("Not X, but Y" / "Not because X, but because Y" / "The problem isn't X, it's Y")
  • Negative listing ("Not a X. Not a Y. A Z.")
  • Dramatic fragmentation ("That's it. That's the thing.")
  • Rhetorical setups ("What if X?", "Here's what I mean:", "Think about it:")
  • False agency ("the culture shifts", "the decision emerges", "the data tells us")
  • Narrator-from-a-distance ("Nobody designed this", "This happens because")
  • Passive voice that hides actors
  • Three-item lists used reflexively
• Lazy extremes (always, never, everyone, nobody) doing vague work.
• Empty adverbs (really, just, literally, genuinely, actually).

Rewrite principles

Adapted from the upstream skills this is built on. The full source is in references/.

1. Cut throat-clearing. State the point. Don't announce it.
2. Active voice. Every sentence needs a human subject doing something. Name the actor.
3. Be specific. No vague declaratives. Name the thing.
4. Put the reader in the room. "You" beats "people". Concrete beats abstract.
5. Vary rhythm. Mix sentence lengths. Two items beat three. Connect short sentences with conjunctions, semicolons, or commas to break parataxis.
6. Trust readers. Skip the softening, the justification, the hand-holding.
7. Cut quotables. If it sounds like a pull-quote, rewrite it.
8. Show, don't describe. "Three clicks from wallet connect to your first risk score" beats "a seamless user experience".
9. Include friction. Real human writing has doubt, mess, specific dates and times.

Apply silently

Don't say "as per the unsloppify rules" or "applying rule X". Just rewrite. The user can see the diff.

References

• references/phrases.md — banned phrases and vocabulary
• references/structures.md — structural patterns
• references/examples.md — before/after rewrites
• references/rubric.md — five-dimension scoring

Credit

Vocabulary and structural rules adapted from MIT-licensed:

• hardikpandya/stop-slop — the five-dimension rubric, structural patterns, business-jargon swaps
• jalaalrd/anti-ai-slop-writing — banned vocabulary list, parataxis rules, accuracy rules

License

MIT