Skip to content

feat: add quality scoring, core principles, and 3 wiki-sourced patterns #94

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Charpup wants to merge 1 commit into
base: main
Choose a base branch
from
Open

feat: add quality scoring, core principles, and 3 wiki-sourced patterns #94

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
74 changes: 74 additions & 0 deletions SKILL.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -461,6 +461,27 @@ Avoiding AI patterns is only half the job. Sterile, voiceless writing is just as
>
> When users hit a slow page, they leave.


### RELIABILITY AND EVIDENCE PATTERNS

30. **Citation Fabrication / Hallucinated Sources**
LLMs sometimes generate citations that look legitimate but are off-topic, have invalid DOIs, or reference non-existent publications. Also watch for: broken external links, `utm_source=` parameters in URLs, named references declared but never used.
> Fix: Verify every citation. Remove unverifiable ones.
- Before: "According to Smith et al. (2023) in the Journal of Digital Innovation..."
- After: [verify the source exists and says what you claim, or remove]

31. **Broken Markup Artifacts**
AI-generated text may contain markup from a different context: markdown syntax in Word documents, `oaicite` tags, `contentReference` spans, or `turn0search0` references.
> Fix: Remove all AI markup artifacts. Use the platform's native formatting.
- Before: "The results were **significant** (see `contentReference[oaicite:0]`)"
- After: "The results were significant (see Table 2)"

32. **Era-Specific Vocabulary Clustering**
AI vocabulary clusters by model era. When you spot one high-frequency AI word, scan the surrounding text — others from the same era tend to cluster in the same paragraph. Finding 3+ AI vocabulary words in one paragraph is a strong signal.
> Fix: When you spot a cluster, rewrite the entire paragraph, not just individual words.

> Patterns 30-32 sourced from [Wikipedia: Signs of AI writing](https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing) and [WikiProject AI Cleanup](https://en.wikipedia.org/wiki/Wikipedia:WikiProject_AI_Cleanup)

---

## Process
Expand Down Expand Up @@ -489,6 +510,59 @@ Provide:
4. A brief summary of changes made (optional, if helpful)


## 5 Core Principles (Quick Reference)

1. **Delete filler** — Remove filler words and emphatic crutches. "It is important to note" means nothing. Delete it.
2. **Break formula** — Avoid binary contrasts ("not just X, but Y"), dramatic segmentation, rhetorical setup. Start with substance.
3. **Vary rhythm** — Mix sentence lengths. Short. Then longer ones that develop the thought. Two items beat three. Vary paragraph endings.
4. **Trust the reader** — State facts directly. Skip softening, excuses, hand-holding. If you wrote "to be sure" or "of course" — delete it.
5. **Kill the quotables** — If a sentence sounds like it belongs on a motivational poster, rewrite it. Real people don't speak in epigrams.

> Adapted from [hardikpandya/stop-slop](https://github.com/hardikpandya/stop-slop)


## Quick Checklist (Pre-Delivery Audit)

- [ ] Three consecutive sentences same length? → Break one up
- [ ] Paragraph ends with short single line? → Vary the ending
- [ ] Em dash before a revelation? → Delete it
- [ ] Explaining a metaphor? → Trust the reader
- [ ] Used "Additionally" / "However" / "Furthermore"? → Consider deletion
- [ ] Rule of three anywhere? → Change to two or four


## Quality Scoring

After rewriting, score your output across 5 dimensions (1-10 each, total /50):

| Dimension | 10 = | 1 = |
|-----------|------|-----|
| **Directness** | Every sentence delivers info immediately | Every paragraph starts with setup |
| **Rhythm** | Mix of short and long, varied endings | Mechanical uniformity |
| **Reader Trust** | Concise, no over-explanation | Hand-holds throughout |
| **Authenticity** | Has voice, opinions, specific observations | Obviously AI, formulaic |
| **Conciseness** | Zero redundancy | Could be cut 40%+ |

**Target: ≥ 40/50 before delivery.**

> Scoring system adapted from [op7418/Humanizer-zh](https://github.com/op7418/Humanizer-zh), originally from [hardikpandya/stop-slop](https://github.com/hardikpandya/stop-slop)


## Detection Confidence Framework

Not all patterns carry equal weight. Individual LOW patterns may be normal human style.

| Confidence | Patterns | Alone? |
|-----------|----------|--------|
| **HIGH** | P1, P3, P5, P6, P7 (clustered), P16, P18, P20, P21, P22, P25, P28, P30, P31, P32 | Strong signal |
| **MEDIUM** | P2, P4, P8, P9, P10, P11, P14, P15, P23, P24, P27, P29 | Suspicious if 2+ |
| **LOW** | P12, P13, P17, P19, P26 | Normal unless clustered |

**Combined Indicator Principle:** 3+ categories matching in the same text = high probability AI origin.

> Framework based on [Wikipedia: Signs of AI writing](https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing)


## Full Example

**Before (AI-sounding):**
Expand Down