No AI Slop: evidence-led editing, packaging and evaluation
No AI Slop explained: editable writing patterns and the limits of detection
Learn what the writing skill edits, what its detection report should contain, and why neither mode proves whether a person or model wrote a text.
What you will learn
- What the project actually supplies
- Editing and detection produce different deliverables
- When it helps and when it needs restraint
Before you start
- A draft whose facts can be checked
- Basic understanding of assistant instructions and plugin scope
Inspect pattern evidence, preserve meaning and distinguish package checks from unmeasured editing outcomes.
Key takeaways
- The deliverable is an instruction package for an existing assistant.
- Detection should cite patterns, not guess authorship.
- Package validation does not establish editing quality.
What the project actually supplies
No AI Slop packages editing instructions for an existing assistant. Its canonical SKILL.md describes minimal edits that preserve meaning, distinctive vocabulary, rhythm and uncertainty. The repository also contains an evaluation checklist and a Python package builder. At commit 000650b, the plugin manifest declares version 1.0.6; there is no separate model in this package.
Consider a release note that says a change demonstrates commitment to innovation but never explains what users can now do. The skill targets that empty commentary and asks for a concrete consequence already supported by the draft. It must not manufacture a customer benefit or performance number simply because a specific sentence sounds better.
Editing and detection produce different deliverables
Editing returns the full revised draft and a short account of changes. Detection instead names a pattern, quotes the relevant line and offers a brief fix without rewriting or assigning an authorship score. A line can match a pattern whether its author is human or a model. Treat the quoted passage as inspectable editorial evidence, not a forensic test.
The README also suggests a satire prompt. That demonstration does not add a third production analysis engine: the canonical skill defines edit and detect workflows. For a team evaluation, use an ordinary draft with known facts rather than a comic example deliberately packed with every prohibited phrase.
When it helps and when it needs restraint
The skill is useful when edits routinely erase an author’s voice or leave generic transitions untouched. It is less suitable as an unquestioned filter over legal quotations, technical terminology or deliberate literary repetition. The rule set contains strong stylistic preferences; whether a particular sentence should change still depends on audience and context.
This series separates instruction behavior from package behavior. We executed seven isolated validation cases against the pinned Python functions, using synthetic files. Those cases reveal a Windows path-separator limitation in package validation, but they say nothing about whether a model preserves humor, improves reading or detects every pattern. No live writing evaluation was run.
Decision guide
| Criterion | Option A | Option B |
|---|---|---|
| Best when | You need predictable behavior and easy auditing | You need adaptive optimization and have reliable telemetry |
| Main risk | May leave performance on the table | Can become difficult to explain or debug |
Implementation steps
- 1
Choose one draft with facts you can verify.
- 2
Decide whether you want findings or a rewritten draft.
- 3
Identify voice traits and passages that must remain unchanged.
Copy-ready example
{
"project": "petergyang/no-ai-slop",
"manifestVersion": "1.0.6",
"mode": "detect",
"authorshipInferenceAllowed": false,
"modelEvaluationExecuted": false
}Frequently asked questions
Can it prove that AI wrote an article?
No. Its detection instructions explicitly prohibit guessing AI authorship.
Does a successful package test validate the writing rules?
No. File and manifest checks do not measure the quality of a model response.
Sources
- No AI Slop / README.mdSource checked 2026-09-14
- No AI Slop / skills/no-ai-slop/SKILL.mdSource checked 2026-09-14
- No AI Slop / .codex-plugin/plugin.jsonSource checked 2026-09-14