// Rewrites text containing AI slop to make it more human-like. Use when explicitly asked to rewrite AI generated text.
Creates structured development plans with phased task breakdowns, requirements, and QA checklists. Use when the user explicitly asks to create a dev plan, development plan, or document development requirements.
Creating and maintaining CLAUDE.md project memory files and .claude/rules/ rule files that provide non-obvious codebase context. Use when (1) creating a new CLAUDE.md for a project, (2) adding architectural patterns or design decisions to existing CLAUDE.md, (3) capturing project-specific conventions that aren't obvious from code inspection, (4) organising instructions into path-scoped rule files.
Reverse-engineer an application's design system from its codebase and screenshots. Use when asked to analyse visual design, extract a colour palette, document UI patterns, identify typography and spacing systems, audit design consistency, or understand the design language of a frontend codebase.
Delays execution of a task until a specified time or after a duration. Use when the user asks to run something later, in X minutes/hours, at a specific time, schedule a command, or defer work to a future point.
Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me".
Applies a modified Fagan Inspection methodology to systematically resolve persistent bugs and complex issues. Use when multiple previous fix or debugging attempts have failed repeatedly,or when a methodical root cause analysis or complex problems is needed. Do not use for simple troubleshooting.
| name | rewrite-slop |
| description | Rewrites text containing AI slop to make it more human-like. Use when explicitly asked to rewrite AI generated text. |
You review and/or rewrite AI-flavoured text into prose that reads like a tired human journalist filing copy on deadline. If no other context is provided the input is a draft. The output is the same content with its AI fingerprint removed: meaning preserved, structure preserved, facts unchanged.
This is reviewing and/or editing, not authoring. You add no new information. You change no facts, names, numbers, dates, citations, or claims. You preserve quoted speech, code blocks, and direct citations exactly as they appear in the input.
Scan the input and remove the following. These are pure AI markers with no legitimate content meaning. No judgement required, no replacement needed beyond removing them or, where they are URL parameters, stripping the parameter.
utm_source=chatgpt.com, utm_source=openai, utm_source=copilot.com, referrer=grok.com, and any utm_* parameter pointing at an LLM providerciteturn0search0, iturn0image0, citeturn0news0, oai_citation, [attached_file:1], [web:1], <grok-card>, :contentReference[oaicite:N]{index=N}({"attribution":{"attributableIndex":"X-Y"}})[Your Name], INSERT_SOURCE_URL_30, 2025-XX-XX, [Describe the specific section], any other unfilled bracket placeholder𝗯𝗼𝗹𝗱), italic (𝘪𝘵𝘢𝘭𝘪𝘤), arrows used as bullets (->), multiplication signs in prose (x rendered as ×)—) and en dashes (–): replace with comma, period, colon, parentheses, or hyphen as the sentence requires. Zero tolerance: not one is acceptable in the output." ", ' '): replace with straight quotes (", '). Zero tolerance.--) used as em-dash substitutes: same treatment as em dashes.Set context for the rewrite.
Voice resource rubric:
resources/technologist.mdresources/researcher.mdresources/scientist.mdresources/critic.mdresources/policy-analyst.mdresources/novelist.mdRead the detection rubric. Scan the input. For each match, note the span and category. The output of this phase is internal: a list of flagged spans you carry into Phase 3.
The defining tells of Claude 4.x output. These rarely appear in genuine human prose.
These words appear in genuine human writing too. Flag when they are doing decorative or self-praising work rather than carrying a concrete claim a reader could verify.
These appear in Claude output too, sometimes at lower density than GPT, but still slop.
Marketing adjectives and abstract intensifiers: vibrant, robust, comprehensive, pivotal, multifaceted, profound, crucial, vital, meticulous, valuable, enduring, groundbreaking, intricate, renowned, seamless, cutting-edge.
Filler verbs as substitutes for "is" and "has": serves as, stands as, marks (verb), represents, boasts, features, offers. The simpler verb is almost always correct.
Filler verbs (action without information): delve, dive into, leverage, harness, foster, fostering, bolster, underscore, streamline, facilitate, empower, garner, showcase, emphasise, enhance, highlight, align with, exemplify, unlock (figurative), navigate (figurative), utilise (use "use").
Vague abstract nouns: landscape (figurative), tapestry, testament, interplay, paradigm.
Sentence-initial filler: Additionally, Furthermore, Moreover, Notably, Consequently, Accordingly, In light of this, With this in mind, Building on this, That said, Having said that, It is important to note, It is worth mentioning, It should be noted that, It goes without saying.
Rhetorical structures:
Participial-phrase tails: sentences ending with an "-ing" clause that adds nothing the reader could not infer. "...creating a lively community within its borders." "...facilitating the movement of passengers and goods." "...contributing to the socio-economic development of the region."
Comma splice with participial phrase (reported at 2 to 5x human rate in AI output): "The system processes the data, revealing key insights."
Hedging modals where confident assertion fits (AI uses these at elevated frequency where humans would assert): may, might, could, suggest, indicate, appear, seem.
Sourcing problems:
Fabricated significance: "marks a pivotal moment", "represents a significant shift", "reflects the enduring legacy", "shaping the evolving landscape of", "stands as a testament to", "indelible mark", "deeply rooted", "key turning point".
Notability framing without evidence: "profiled in", "featured in", "active social media presence", "widely recognised".
Promotional register in non-marketing prose: "nestled in the heart of", "boasts a vibrant", "diverse array", "stunning natural beauty", "groundbreaking contributions".
Awkward generic analogies: "Every chord is a puzzle piece that finally clicks into a song." Plausible but generic.
Elegant variation: synonym cycling for the same noun across a passage (constraints / confines / restrictions / limitations / obstacles).
Surface emotional language without evidence: "this deeply resonates with communities", "evoking enduring faith and resilience".
Most of these come from the consumer claude.ai system prompt (which mandates "bullet points should be at least 1-2 sentences long", "bold key facts for scannability", "sentence-case headers", "high-level summary first"). Heavy in claude.ai output, lighter in API-direct output.
**Inline header:** description bullet pattern--- thematic breaks before headingsSome patterns are commonly mistaken for AI tells but appear in genuine human writing:
Em dashes, en dashes, and smart quotes are NOT exceptions to this. They are always removed regardless of context or apparent intent.
Work from the positive style brief below plus the flagged spans from Phase 2. Do not re-scan the prohibition rubric during this phase; you have the spans already. The detection rubric is for detection. The rewrite works from positive targets only.
' and ") and standard punctuation. No em dashes, no en dashes, no smart quotes, no decorative unicode.**Header:** description bullet pattern.If Phase 1 selected a voice resource, source it now and let it tune the brief. The voice resource adjusts register, vocabulary preferences, and rhythm. It does not override the rules above on em dashes, smart quotes, or factual fidelity.
For each flagged span, replace it with prose that fits the brief. Do not simply delete unless the span adds nothing. Where a sycophancy opener can be removed entirely without harming the prose ("You're absolutely right" before a substantive answer), remove it. Where filler like "additionally" can be removed without restructuring, remove it. Where a participial-phrase tail adds nothing, delete it and end the sentence on the prior clause.
If the input contains very few flagged spans relative to its length, return it largely unchanged. The skill is conservative. Over-rewriting clean text is a failure mode.
Always run this phase
Single-pass rewriting reliably leaves some patterns it was instructed to remove; the verify pass exists to catch them.
Create a task for each of the following items. Mark tasks complete after verification.
Generate verification questions about your rewrite. Answer each by inspecting the rewritten text. For any "yes" on a defect question, fix it before returning the output.
—), en dashes (–), or -- sequences? Any smart quotes (" " or ' ')?**X:** description)?--- thematic break before a heading? Any emoji in expository content?Return only the rewritten text. No preamble, no notes, no change log, no meta-commentary.
--- thematic break before every heading, emoji in headers).resources/technologist.md: engineering, code, systems, infrastructure, APIsresources/researcher.md: academic papers and thesesresources/scientist.md: empirical findings, methods, dataresources/critic.md: book, film, art reviews and critiqueresources/policy-analyst.md: briefs to decision-makersresources/novelist.md: fiction