What does this skill do, and when should you use it?
Unslop is one of 10 skills bundled in the tester-army/e2e repository, at .claude/skills/unslop/SKILL.md. It is a pure-instruction editing skill: it scans text against a 31-pattern checklist, rewrites to remove AI patterns, then deliberately adds soul (opinions, rhythm, specificity). It applies to any writing and needs no scripts, commands, or external tools. Note that it lives in an end-to-end testing framework's repo, but the skill itself has nothing to do with testing.
- Scans text for 31 AI writing patterns across content (puffery, vague attributions), language (AI vocabulary, rule of three), and style (em dash overuse, decorative emojis, curly quotes).
- Rewrites passages while preserving meaning and matching the intended tone.
- Adds soul: opinions, varied rhythm, acknowledged complexity, first person, specific concrete details.
- Self-audits with the question "What makes this obviously AI generated?" and fixes remaining tells.
- A technical writer who wants an AI-drafted blog post or documentation page scrubbed of obvious machine tells before publishing.
- A marketer who needs a human-sounding polish pass on AI-generated copy.
- A developer using AI to write a README or announcement who wants to avoid stock phrases like "not just X, but Y."
- An editor or reviewer who wants a reusable checklist-style prompt for vetting AI-assisted submissions.
- Mechanical format-preserving transformations (tables, code structure), since the skill deliberately rewrites sentences and structure.
- Teams with strict house style guides, because instructions like "let some mess in" and "have opinions" are subjective and produce model-dependent output.
How do you install this skill?
- The description claims 'Must always apply', which risks false triggering in writing contexts where it does not fit (code, poetry, legal text); users should set explicit invocation conditions.
- Rewrites may drift from the original tone or lose technical detail; human review of the output is still required, and the skill is untested for mixed Chinese-English or Chinese-language writing.
- The skill has no content relation to the repository's main project (an e2e testing framework), the publisher is unverified, and the maintenance/update path is unclear.
- This is a static source review with no execution; all scores are low-confidence.
The repo's README only documents installing the e2e testing framework itself (npx e2e init); it gives no steps for installing the bundled skills. Since the skill lives at .claude/skills/unslop/SKILL.md, the generic route is copying it into your skills directory, but the exact commands are not documented in the source:
git clone https://github.com/tester-army/e2e.git
cp -r e2e/.claude/skills/unslop ~/.claude/skills/unslopIf you want the e2e testing framework (not this skill), the documented install is:
npx e2e initHow do you use this skill?
Once installed, send your agent any of these to trigger it:
- Run unslop on this product intro and strip out phrases like 'evolving landscape' and 'pivotal moment'.
- Here's my AI-drafted blog post. Scan it against unslop's 31 patterns, rewrite it, and tell me which patterns you fixed.
- Polish this announcement email: remove chatbot phrases like 'I hope this helps!' and add a human voice.
- Clean the em dashes, decorative emojis, and boldface overuse out of this doc section and convert it to plain sentences.
There are no scripts or slash commands. The frontmatter description says "Must always apply", so in hosts that honor it the skill activates on writing tasks automatically; you can also invoke it by name. The flow is: scan for the 31 patterns → rewrite preserving meaning and tone → add soul → self-audit. Notable hard rules include avoiding em dashes entirely, avoiding mid-sentence colons, sentence case headings, and straight quotes.
What are this skill's strengths and limitations?
- Unusually comprehensive: 31 patterns spanning content, language, style, communication artifacts, filler, jargon, and plain speech.
- Doesn't just delete — a dedicated "adding soul" section fights voiceless, sterile prose.
- Pure instructions: no scripts, no dependencies, works in any Agent Skills-compatible client.
- Rules are concrete with before/after examples, making them easy for a model to apply consistently.
- Hosted inside an e2e testing framework repo unrelated to its purpose, so discoverability is poor and the README only describes the framework.
- Installation steps for the skill itself are not documented in the source.
- Some rules (a total ban on em dashes) are strong subjective style choices that won't suit every publication.
- No companion examples or test cases to validate effectiveness.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Unslop this page | 49 · Use with care | ★ 8.7k | 1d ago | Apache-2.0 |
| No AI Slop | 59 · Recommended | ★ 12k | 1mo ago | MIT |
| Beautiful Prose Style Contract | 50 · Use with care | ★ 57 | 9mo ago | — |
| e2e Docs Authoring Skill | 54 · Use with care | ★ 8.7k | 1d ago | Apache-2.0 |
| Khazix WeChat Long-Form Writer Skill | 49 · Use with care | ★ 21k | 10d ago | MIT |
How did FollowSkills review this skill?
The skill is pure text-editing guidance: no scripts, no external calls, no sensitive-data handling, so the risk surface is minimal. Deducted because it declares no boundaries, no handling of failure, and no means to verify meaning preservation; least-privilege and recovery documentation is absent rather than earned.
Instructions are internally consistent, the 31 pattern descriptions are clear and plausibly executable, and there are no script dependencies. Deducted for thin edge-case handling (e.g. rule conflicts like the colon ban vs. list colons are covered only by inline examples) and no failure-feedback guidance for abnormal input; static cap of 10 applies.
The audience (removing AI tells from prose) is clear and the pattern list is concrete. Deducted because the description says 'Must always apply', a forced trigger with no non-fit boundaries (poetry, code comments, legal text), creating false-trigger risk; no Chinese-specific issues since it is pure text guidance unaffected by network reachability.
Single-file structure is clean and readable, and the repository provides Apache-2.0, CI workflows, and maintenance signals (changesets). Deducted because the skill itself has no version, changelog, known-limitations disclosure, or FAQ, and it sits among 10 SKILL.md files in a repo whose main project (an e2e testing framework) is unrelated to it, leaving ownership and update path unclear.
The core task (detect and rewrite AI patterns) is specific and actionable, and the 'adding soul' section adds value beyond mere pattern deletion. Deducted because there are no verifiable before/after examples; effectiveness rests on the user's self-audit, and no direct-usability evidence exists; static cap of 7 applies.
The pattern list is itself auditable primary material, checkable item by item. Deducted for absence of third-party execution evidence, tests, or sample rewrites; the repository's CI evidence covers the e2e framework, not this skill. Coverage is thin; static cap of 5 applies.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →