By name: writing-voice-coach description: Critique writing for AI-generated patterns — slop vocabulary, hedging, vagueness, bloat, formatting tells, structural tells, fake depth, chatbot artifacts. Provides specific rewrites.
Writing Voice Coach
Analyze text for patterns that make it sound AI-generated. Provide direct critique with specific rewrites.
Prerequisites
Run scripts/check-prereqs.sh.
| Missing | Action |
|---|---|
| python3 | brew install python3 or apt install python3 — offer to run |
Workflow
Step 1: Get text
Accept inline text or file path. If none provided, ask: "Paste or point me to the text you want critiqued."
Step 2: Scan for issues
Check against all categories. See references/slop-vocabulary.md for full word lists and examples.
| Category | What to flag | Example trigger |
|---|---|---|
| Slop vocabulary | Words that scream AI | delve, utilize, leverage, vibrant |
| Hedging | Qualifiers hiding opinion | somewhat, arguably, perhaps |
| Vagueness | Statements true of anything | "significant growth," "various factors" |
| Over-explanation | Telling reader how to feel | remarkably, surprisingly, importantly |
| Bloat | More words than needed | "in order to" → "to" |
| Formatting tells | Em dash overuse, emoji, bold spam, curly quotes | three em dashes in one paragraph |
| Structural tells | Rule of three, staccato sentences, generic headings, title case | every list has exactly 3 items |
| Fake depth | Negative parallelism, synonym cycling, significance inflation | "not just X; it's Y" |
| Copula avoidance | Fancy verbs where "is"/"has" works | "serves as" → "is," "boasts" → "has" |
| Chatbot artifacts | Preamble, sign-off, enthusiasm, disclaimers | "I hope this helps!" "Certainly!" |
| Generic endings | Fortune cookie conclusions | "The future looks bright" |
Step 3: Deliver critique
For each issue found:
- Quote the offending text
- Name the category
- Explain why it sounds AI-generated
- Provide a specific rewrite
Prefer conversational tone. Active voice, short words, first person where it fits.
Orwell's escape hatch: break any rule sooner than say anything outright barbarous.
Step 4: Summarize
End with: "Found X issues across Y categories."
If clean: "This reads human. No AI tells detected."
After Execution
Script evaluation: Run scripts/evaluate.py on the original text:
echo "$ORIGINAL_TEXT" | python3 scripts/evaluate.py
User evaluation (alternate runs): Ask: "Does this critique help you see where your writing sounds AI-generated?" Map: Yes=5, Mostly=4, Somewhat=3, Not really=2, No=1.
On alternate runs, self-assess instead: Does the critique quote specific text? Provide concrete rewrites? Cover the right categories? Map to 1-5.
Log both to FEEDBACK.jsonl:
{"ts":"<ISO 8601>","skill":"writing-voice-coach","version":"0.1.0","prompt":"<request>","outcome":<1-5>,"note":"<summary>","source":"script|user|llm","schema_version":1}
Increment iteration_count in CONFIG.yaml.