A skill file for removing AI tells from prose
13.6k
Stars
958
Forks
37
Open issues
2
Contributors
AI Analysis
Stop Slop is a skill/instruction set for large language models to identify and remove common AI-generated writing patterns—phrases, structures, and rhythms typical of LLM output. It is purpose-built for users (writers, editors, content creators, LLM operators) who want to sanitize AI-assisted or AI-generated prose to sound more human and authentic. It does not assist general software development and is not intended for audiences seeking traditional developer tooling.
Inferred from signals mentioned in the README (tests, CI, type safety) — not a review of the actual code.
AI's overall editorial judgment — not an average of the bars above, can weigh other factors too.
Prompt skill file that strips common AI writing patterns from LLM output prose
stop-slop is a structured skill file (SKILL.md plus reference lists) designed to be injected into Claude or other LLMs to detect and remove telltale AI writing patterns: filler phrases, structural clichés, passive voice, em-dash overuse, and metronomic rhythm. It targets writers, content teams, and developers who use LLMs to draft prose but want output that reads as authored rather than generated. It is not an application or library — it is a reusable prompt artifact. Its value is in the curated lists and scoring rubric it provides, not in any executable code.
Created in January 2026 by product designer Hardik Pandya, apparently as a personal tool shared publicly. It emerged alongside the broader 'AI slop' discourse that intensified in 2025 as LLM-generated text became pervasive and recognizable.
Reached ~11,900 stars by mid-2026, with 595 stars in the last 7 days suggesting an ongoing viral cycle — likely driven by social media shares, newsletter mentions, or influencer pickup. The fork count of 833 indicates genuine reuse intent rather than passive starring. Growth appears episodic rather than steady, consistent with a content-type repo that spreads through writing and AI communities.
Adoption not verified in formal production environments. The fork count (833) and star velocity (595 in 7 days) suggest active interest and personal reuse, but there is no documented case study, integration, or organizational deployment referenced in the README. Individual writers and developers are likely the primary users, with adoption occurring informally.
Appears to be a flat collection of Markdown files: a core SKILL.md with rules, and three reference files (phrases.md, structures.md, examples.md). No programming language is used. Likely designed as a copy-paste or file-upload artifact rather than an importable package. The structure is simple and human-readable by design.
not documented in README — no automated tests are applicable to a prompt-only project of this type
Last push was 2026-03-17, roughly 3 months before the evaluation date. For a static prompt file, this is not necessarily a concern — the content may simply be stable. However, there is no documented versioning, changelog, or issue triage visible from metadata. The project may be functionally complete as-is, or it may have stalled; the distinction cannot be determined from available data.
ADOPT IF: you regularly use Claude or similar LLMs to draft prose and want a reusable, opinionated style guide embedded in your workflow without building custom tooling. AVOID IF: you need programmatic, automated, or CI-integrated AI text detection — this is a manual prompt artifact, not a library or API. MONITOR IF: you are evaluating prompt-as-skill patterns more broadly and want to see whether this specific vocabulary of 'AI tells' proves durable as LLM writing styles evolve.
Independent dimensions
Mainstream potential
3/10
Technical importance
4/10
Adoption evidence
3/10
- The list of banned phrases and patterns may become outdated as LLM outputs evolve — what counts as an 'AI tell' in early 2026 may differ significantly within 12–18 months.
- Effectiveness depends entirely on the LLM following the instructions faithfully; prompt compliance varies by model, version, and context length, and cannot be guaranteed.
- No programmatic enforcement exists — a user who forgets to include SKILL.md gets no benefit, making adoption fragile in team or automated workflows.
- The scoring rubric (1–10 per dimension) is subjective and not calibrated against any external benchmark, which limits its utility as a quality gate.
- Maintenance appears to have slowed since March 2026; if the author does not continue updating the phrase and structure lists, the resource may drift out of relevance.
Likely to remain a useful personal reference for individual writers and LLM power users, but unlikely to evolve into a widely standardized tool without programmatic integration or active community curation.
Newsletter
Get analyses like this every Monday
Free weekly digest of the most interesting open-source discoveries.
Languages
No language breakdown available.
Information
- License
- MIT
- Last updated
- 4mo ago
- Created
- 6mo ago
- Analyzed with
- anthropic/claude-haiku-4-5
Stars over time
Contributors over time
Top 100 contributors only — repos with more will plateau at 100.
Open issues
Top contributors
Recent releases
No releases published yet.
Similar repos
conorbronsdon/avoid-ai-writing
A portable writing skill for AI agents (Claude Code, OpenClaw, Hermes) that...
blader/humanizer
Humanizer is a portable agent skill that removes AI-generated writing patterns...
nidhinjs/prompt-master
Prompt Master is a Claude skill that optimizes prompts for any AI tool by...
hexiecs/talk-normal
talk-normal is a system prompt designed to make any LLM produce concise, direct...
| Repository | Stars | Week Δ | Language | Score | Updated |
|---|---|---|---|---|---|
|
hardikpandya/stop-slop
this
|
13.6k | +389 | — | 7/10 | 4mo ago |
|
conorbronsdon/avoid-ai-writing
|
2.3k | — | JavaScript | 7/10 | 2d ago |
|
blader/humanizer
|
28.5k | — | — | 7/10 | 2w ago |
|
nidhinjs/prompt-master
|
10.4k | — | — | 7/10 | 4w ago |
|
hexiecs/talk-normal
|
1.8k | — | Shell | 7/10 | 3mo ago |
 addyosmani/agent-skills
|
76.5k | — | JavaScript | 8/10 | 10h ago |
Appears to target the same 'de-AI-ify text' problem but has more than double the stars, suggesting earlier or broader community traction. Without a README comparison it is unclear whether it is also a prompt file or an application.
A broader skill-file collection by a high-profile author (Google Chrome team). Likely covers many skill types beyond prose editing, making it less focused but far more adopted. stop-slop is narrower and more opinionated.
Likely a collection associated with a prominent AI figure, explaining its outsized star count. Not directly comparable — probably covers technical/coding skills rather than prose style.
Appears to be a general prompt engineering resource. Less specialized than stop-slop; serves a different audience (prompt engineers vs. writers).
A Python-based tool from Microsoft, suggesting it is a programmatic skill optimization framework rather than a prompt file. Different abstraction level and audience; not a direct substitute.