Skip to content

iam25th1/no-slop

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

no-slop

A Claude skill that strips the behaviors users complain about most: sycophancy, preamble, excessive hedging, unsolicited moralizing, over-structuring, closing offers, and AI tics ("delve", "let me break this down", em dashes, "I hope this helps!").

Built from recurring public feedback on where Claude feels annoying.


before / after

prompt: how do python generators work?

without no-slop:

Great question! Python generators are a really powerful feature of the language. Let me break this down for you. A generator is a special type of iterator that...

with no-slop:

A generator is a function that produces values one at a time using yield instead of return. Each call to next() runs the function until the next yield, hands back that value, and pauses the function's state.

12 more before/after pairs in evals/examples.md.


install

Pick whichever fits your setup:

Anthropic API (recommended for most cases): Paste the contents of SKILL-compact.md into your system prompt. It is 2.2kb so it will not eat your context budget.

Claude Code:

git clone https://github.com/iam25th1/no-slop.git ~/.claude/skills/no-slop

claude.ai Projects: Paste the contents of SKILL.md into your Project's custom instructions.

Cursor / other IDE AI tools with system prompt config: Paste SKILL-compact.md into the system prompt field.


what's in the box

file what it is
SKILL.md full version, 16 rules with examples and a pre-send checklist, 6.8kb
SKILL-compact.md token-tight version, 2.2kb, all 16 rules preserved
evals/evals.json 12 test prompts in anthropic skill-creator eval schema
evals/examples.md before/after pairs for eyeballing whether it is working

what it fixes

complaint rule
launching into solutions when someone was venting 3
"great question! let me break this down..." 1, 2
ending every response with "would you like me to..." 9
stacked hedges ("it's worth noting... keep in mind that...") 5
"both have merit, it depends" when asked A or B 6
agreeing with flawed premises 7
bullet lists for a conversational question 4
explaining git to someone debugging a merge conflict 10
unsolicited ethics lectures on technical questions 13
"delve", "dive into", em dashes, "I hope this helps!" 12
"I'll be concise" followed by a paragraph 8
fake hedging instead of "I don't know" 14
refusing then offering an unrelated weaker task 15

what it does NOT override

  • Safety rules (harm, child safety, weapons, etc.)
  • Explicit user instructions
  • Legal or medical contexts where hedging is genuinely protective
  • Creative writing where voice and flourish are the point
  • Formal documents where register matters

User preferences always win.


testing the skill

Run the 12 prompts from evals/examples.md twice, once with the skill loaded and once without. Diff the responses. If a "with skill" response still looks like the "slop" column in examples.md, that rule is not landing and needs sharper wording.

For the quantitative route: the JSON in evals/evals.json is compatible with anthropic's skill-creator eval pipeline if you want automated grading.


contributing

Spotted a rule that is wrong, a tic that should be added, or a case where the skill over-corrects?

  1. Fork
  2. Add a failing test case to evals/examples.md showing the problem
  3. Propose the fix (new rule, edited rule, removed rule)
  4. PR

Rules that cannot be demonstrated with a before/after example probably do not belong in the skill.


credit

Built by @25thprmr based on recurring patterns in public Claude feedback (reddit, x, HN, anthropic forums).

license

MIT

About

Strip AI-slop from Claude responses. 16 rules, ready to drop into API system prompts or Claude Code

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors