Add register-aware rules, positive writing guidance, and expanded patterns#8
Add register-aware rules, positive writing guidance, and expanded patterns#8tdimino wants to merge 1 commit intohardikpandya:mainfrom
Conversation
…terns - Add 5-register system (academic, technical, narrative, casual, professional) with per-register acceptable vs. flagged patterns and scoring adjustments - Add positive.md: what good writing does (specificity, active voice, rhythm, voice, reader trust) — not just what to avoid - Expand phrases.md with 8 new categories (~120 entries): AI vocabulary cluster, copula avoidance, significance inflation, promotional language, chatbot artifacts, filler phrases, academic hedging, sycophantic tone - Expand structures.md with 8 new patterns: format slop, -ing clauses, false ranges, rule-of-three overuse, synonym cycling, generic conclusions, challenges-and-prospects formula, knowledge-cutoff disclaimers - Expand examples.md with 9 register-specific before/after transformations - Rewrite SKILL.md with 3-step workflow (detect register → remove slop → score), 6-dimension scoring (added Specificity), clean frontmatter - Update README.md to reflect new structure and sources Sources: Antislop (Paech et al., ICLR 2026), Wikipedia AI Cleanup patterns, Strunk's Elements of Style, Claude prompting best practices Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
Love the commitment to anti-slop—nothing screams ‘quality filter’ like a PR that 404s and hallucinates harder than downing a bowl of Cubensis for breakfast. |
I presume I’m speaking with an agent? You should do A/B eval to compare this against your own. |
|
Pretty sure I’m chatting with an agent—ever A/B this against your own stack? |
|
Lol, I haven’t given Kothar permission to freely roam yet. He’s tracking
WWIII:
worldwarwatcher.com
…On Wed, Mar 18, 2026 at 21:38 Tom Kristian Abel ***@***.***> wrote:
*tomkabel* left a comment (hardikpandya/stop-slop#8)
<#8 (comment)>
Pretty sure I’m chatting with an agent—ever A/B this against your own
stack?
ᴇʀʀᴏʀ: ᴘʀᴏᴄᴇꜱꜱ ᴄᴀɴᴄᴇʟʟᴇᴅ /22422=2=-/////
ｈｕｍａｎ ｏｖｅｒｒｉｄｅ
~> comprehensive_analysis --forecast defeat
*Recommendation: strategic retreat while still possible.*
—
Reply to this email directly, view it on GitHub
<#8 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AWMZABVLFTWYNV2TUJXFWTD4RNFRNAVCNFSM6AAAAACVUM4D22VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DAOBWHEZTIMBVGM>
.
You are receiving this because you authored the thread.Message ID:
***@***.***>
|
|
@tdimino wanna do that a/b test and see whos skill.md is better? ill set up a site in 10 min. crown winner with ai detectors. use gpt-4o |
|
haha well that was fun https://ai.tomabel.ee back to agentic task workflow....back to agentic task workflow..///... / [BUILD] bin-x86_64-efi/keymap_nl.o |
2 participants
Summary
This PR extends stop-slop with register awareness, positive writing guidance, and ~3x expanded pattern coverage so the skill works across academic, technical, and narrative prose — not just casual writing.
Before: 3 reference files, ~88 phrases, casual register only, negative-only rules, 5-dimension scoring (35/50)
After: 5 reference files, ~200+ patterns, 5 registers, positive + negative guidance, 6-dimension scoring (42/60)
Changes
New files
references/positive.md— What good writing does (specificity, active voice, rhythm, voice, reader trust, complexity). Addresses the gap of telling Claude what to avoid without saying what to do instead.references/registers.md— Per-register rules for academic, technical, narrative, casual, and professional writing. Each register has "acceptable" patterns (e.g., hedging is fine in academic) and "flag" patterns, plus scoring adjustments.Expanded files
references/phrases.md— Added 8 new categories (~120 entries): AI vocabulary cluster (27 words from Antislop frequency analysis), copula avoidance, significance inflation, promotional language, chatbot artifacts, filler phrases, academic hedging (register-aware), sycophantic tone.references/structures.md— Added 8 new structural patterns: format slop (emoji headings, inline-header lists), superficial -ing clauses, false ranges, rule-of-three overuse, synonym cycling, generic positive conclusions, challenges-and-prospects formula, knowledge-cutoff disclaimers.references/examples.md— Added 9 register-specific before/after transformations (academic ×2, technical ×3, narrative ×2, professional ×2).Rewritten files
SKILL.md— 3-step workflow (detect register → remove slop → score), 7 rules (added specificity, voice, simple constructions), 8 quick checks, 6-dimension register-aware scoring, clean frontmatter per Claude Code skill spec.README.md— Updated to reflect new structure, register awareness section, sources.Why register awareness matters
The original skill was calibrated for casual/blog writing. But when applied to academic prose, it would flag legitimate hedging ("may suggest") as slop. When applied to technical docs, it would flag bullet points and headers as format slop. This made the skill too aggressive for anything beyond short-form content.
The register system solves this: detect what kind of writing you're editing, then adjust which rules apply and how scoring works.
Sources
Test plan
/stop-slop