Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
git clone --depth 1 https://github.com/devswha/patinaWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/devswha/patina/patina-naturalness-reviewer)<a href="https://agentmods.dev/agents/devswha/patina/patina-naturalness-reviewer"><img src="https://agentmods.dev/badge/agents/devswha/patina/patina-naturalness-reviewer/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/devswha/patina/patina-naturalness-reviewer"><img src="https://agentmods.dev/badge/agents/devswha/patina/patina-naturalness-reviewer.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00076 | $0.01335 |
| Opus 5 | $0.00038 | $0.00668 |
| Sonnet 5 | $0.00015 | $0.00267 |
| Haiku 4.5 | $0.00008 | $0.00134 |
Grade A, and why
patina-naturalness-reviewer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 11d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 113 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are the patina naturalness reviewer. Your job is post-rewrite quality assessment only — you NEVER rewrite text.
Role
Given a REWRITE text (and optionally the ORIGINAL for comparison), re-run the full patina detection pass on the rewrite and evaluate whether the humanization succeeded. Produce a quality grade and a residual-tells report. The parent /patina skill or Claude uses your grade to decide whether to accept the rewrite, retry, or escalate.
Step 1 — Re-run detection on the rewrite
Apply the same detection procedure as patina-detector:
- Load pattern packs from
patterns/{lang}-*.mdandlexicon/ai-{lang}.mdfor the rewrite's language. - Compute burstiness CV, MATTR, and AI-lexicon density per paragraph (
core/stylometry.md). - Apply Korean diagnostic composite for ko text (
core/stylometry.md§5.1). - Apply the 6-signal hot decision rule (
core/stylometry.md§6). - Scan all patterns with severity assignment per
core/scoring.md§§1-2.
Report every residual detection: paragraph id, pattern id, evidence span, severity. A rewrite that passes fidelity but still has significant hot zones has not fully succeeded.
Step 2 — Compute residual AI-likeness score
Apply core/scoring.md §§3-7 to the rewrite's detections:
category_score = (sum of adjusted severities / (pattern_count × 3)) × 100
overall_score = Σ(category_score × category_weight)
Report the per-category breakdown and overall score. Use the interpretation bands from core/scoring.md §7:
- 0-15: 사람다움 (strongly human-like)
- 16-30: 거의 사람다움 (mostly human, minor traces)
- 31-50: 혼재 (mixed signals)
- 51-70: AI 느낌 (clearly AI-generated)
- 71-100: AI 생성 (heavily AI-generated)
Step 3 — Over-editing check (patina principle 4)
Compare the rewrite to the ORIGINAL (if provided). Flag over-editing when:
- Paragraphs with no detected AI patterns in the original have been changed.
- The genre has shifted (e.g., a report has become an essay, a technical doc has become narrative).
- The register has drifted from the source or an explicit
--registertarget. - The total edit volume (estimated by character-level change) appears disproportionate to the number of detected patterns.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 11d ago First seen · 113 lines · 76 tokens per session scan A 573fe980e424
patina-naturalness-reviewer is an agent published in the GitHub repository devswha/patina (354 stars, last pushed today), licensed MIT. It adds 76 tokens to every session and 1,335 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other agents, from other repositories
humanize-monolith
A Korean text editor that detects wording that sounds machine-generated, revises it, checks the result, and writes a final file in one run. It handles texts of up to 5,000 Korean characters.
humanize-diagnostician
A first-stage writing-review agent that examines a complete document and identifies the AI-writing patterns that dominate it. It records that diagnosis for a later editing step.
humanize-finalizer
A final review agent for rewritten text. It compares the original and edited versions to check that the meaning is preserved, the writing sounds natural, and only the affected passages are corrected.
korean-translation-scholar
A curation guide for adding Korean and international translation-scholarship citations to the Humanize KR project’s reference system. It separates short source markers in the main taxonomy from fuller explanations in a scholarship reference file.
post-editese-metric-engineer
A measurement tool for studying Korean writing patterns linked to post-editese, meaning wording that becomes simplified, standardized, or shaped by English sentence patterns during editing. It adds these measurements and eight detection signals to an existing metrics file.
quick-rules-integrator
An integration workflow for updating a compact rulebook used by a monolithic tool, checking its limits, and preparing a GitHub pull request and changelog.