Menu
Remove AI writing patterns from prose. Use when drafting, editing, or reviewing text to eliminate predictable AI tells.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Rebase all non-pushed commits and uncommitted changes into clean logical feature commits.
Make prose sound like Thomas, not like a generic technical writer. Use whenever drafting, editing, or reviewing any blog post, page, or bio for tschuehly.de — even a small edit. Loads an empirical voice profile extracted from real pre-AI-era posts.
Run the skill-scout loop — scan the next batch of unscanned JVM-conference rosters for speaker-created AI skills and apply results to the CSV store via the overnight Workflow. Use when the user says "run skill-scout", "continue the skill-scout loop", "scan more conferences", or wants to grow the skill candidate list.
Write a new Kotlin test, or modernize an existing one, using Kotest matchers and idiomatic Kotlin test style — backticked names, apply/assertSoftly blocks, and existing object mothers and helpers. Use when adding a test for behavior that does not yet have one, or when improving a Kotlin test that still uses JUnit assertions or AssertJ chains. Assumes project test infrastructure exists; flag missing factories or helpers rather than building them here.
Implement a feature or fix a bug using TDD — write failing test first, then implement, refactor, verify, beautify. Use for any single code change that should be test-driven.
Bootstrap pitest via the info.solidsoft.pitest Gradle plugin with Kotlin-sane defaults, run mutation tests scoped to changed classes, interpret surviving mutants from mutations.xml, triage likely-equivalent mutants out of the kill queue, and drive a kill-survivor workflow. Use when the user asks to add mutation testing, check mutation score, investigate surviving mutants, or strengthen test quality beyond line coverage.
| name | stop-slop |
| description | Remove AI writing patterns from prose. Use when drafting, editing, or reviewing text to eliminate predictable AI tells. |
| metadata | {"trigger":"Writing prose, editing drafts, reviewing content for AI patterns","author":"Hardik Pandya (https://hvpandya.com)"} |
Eliminate predictable AI writing patterns from prose.
Cut filler phrases and AI-marker words. Remove throat-clearing openers and emphasis crutches; replace excess LLM vocabulary (delve, underscore, foster, pivotal). See references/phrases.md.
Break formulaic structures. Avoid binary contrasts, dramatic fragmentation, rhetorical setups. See references/structures.md.
Vary rhythm. Mix sentence lengths. Two items beat three. End paragraphs differently.
Trust readers. State facts directly. Skip softening, justification, hand-holding.
Cut quotables. If it sounds like a pull-quote, rewrite it.
No em-dashes. Replace every em-dash (—), whether before a reveal, around an aside, or anywhere else, with a colon (introducing a list or expansion), commas or parentheses (around an aside), or a full stop. The em-dash is one of the strongest AI tells; the target in prose is zero. Keep dashes only in numeric ranges (8–25, an en-dash) and inside code.
Before delivering prose:
—) at all? Replace it with a colon, commas, parentheses, or a full stop. Zero em-dashes in prose.Before scoring, run one reflective pass:
This catches patterns the rules miss. The rubric below measures; this pass thinks.
Rate 1-10 on each dimension:
| Dimension | Question |
|---|---|
| Directness | Statements or announcements? |
| Rhythm | Varied or metronomic? |
| Trust | Respects reader intelligence? |
| Authenticity | Sounds human? |
| Density | Anything cuttable? |
Below 35/50: revise.
Use this when reviewing a draft cold, from a separate session or subagent that did not write it. Report only. Do not edit. The cold read is the point; rewriting throws it away.
Output, in order:
Tells found: a table, one row per concrete instance:
| Tell | Location (line / quote) | Fix |
|---|
Scan for em-dashes explicitly and flag every em-dash (—) in prose as its own row; the target is zero.
Rubric score: the five dimensions above, each 1-10, plus the total.
Verdict: pass (≥35/50, no Tier-1 LLM vocabulary, and no em-dashes in prose) or revise.
Quoting the exact line keeps the fixer honest. The writer applies the fixes; the auditor never touches the prose.
See references/examples.md for before/after transformations.
MIT