Vladimir-Human/humanizer-ru#dsh
Cleans AI traces from Russian text: finds chatbot copy-paste artefacts (ChatGPT, Gemini, Grok, Perplexity, DeepSeek), strips hidden text marks (zero-width characters, invisible layout; shadow scan catches markers split by invisibles), and rewrites the text into natural prose on request; 40 regex markers, 38 with full evidence-registry entries; 156 gates, 530 tests; offline, text-only bundle (binary metadata removal lives in scripts/).
Install
This plugin lives in a subdirectory of a larger repository, and `dsh plugin add` cannot reach it — it forwards to pnpm, which reads everything after `#` as a git branch or commit. Clone the repository and add the plugin through DSH's plugin panel using a repository source, or ask the author to publish it to npm.
Due diligence
What it is — Tool to detect and rewrite AI traces in Russian text into natural prose on request.
Who it is for — If you need to remove AI generation traces from Russian text when analyzing or editing content in DeepSeek Harness, this plugin is suitable. Users who require a plugin that supports interactive interfaces or code execution may find it unsuitable, as this is an offline text-only solution.
Watch out — No obvious pitfalls found. Static checks reported no issues. Sandbox testing has not been conducted.
The verdict — I would install it because it offers an offline text-only solution with evidence registry, but the premise is that my tasks involve processing Russian AI-generated text.
Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.
humanizer-ru is a DeepSeek Harness plugin that detects and removes traces of machine-generated Russian prose, then rewrites supplied text into more natural language when requested.
It works as an agent skill rather than a runtime service. The agent loads SKILL.md for the analysis or editing task, together with the relevant files from references/. The skill defines 38 patterns—25 basic and 13 extensions—and 39 checkable regular-expression markers in classes A and B. Its evidence registry records support for 37 of the 39 markers, and the repository runs its checks in CI. The agent receives a prepared Russian text fragment, identifies the markers, and performs a meaning-preserving rewrite without adding facts that were not present. PERSONA.md is separate: it supplies conversational tone guidance and is not a text-checking reference.
This plugin is intended for agents editing Russian-language copy where avoiding false positives matters more than catching every possible signal. It is not a detector-bypass tool and does not tune text for a detector’s verdict; use it when the goal is natural prose, not a green score. It is also not a general conversational persona: loading SKILL.md into a chat client’s system prompt is specifically discouraged because it adds token cost without making dialogue more natural. The working bundle is text-only and does not execute code when activated; the repository’s Python scripts are for manual or GitHub Actions testing.
Provide humanizer-ru with a prepared Russian text fragment. The agent uses SKILL.md to analyse it and, when asked, rewrite it into more natural prose while preserving the source meaning. It removes stylistic clichés and chatbot copy-and-paste residue, but does not add facts. The documented examples remove wording such as promotional labels, inflated claims and stock descriptions while retaining the concrete features already present in the input.
The detection material contains 38 patterns: 25 basic patterns and 13 extensions. These correspond to 39 checkable regex markers in classes A and B. The repository maintains an evidence registry for the markers, with evidence recorded for 37 of 39, and checks the patterns in CI.
The working skill is made up of text files. Load SKILL.md for a text-analysis or editing task, along with the required reference files in references/. PERSONA.md contains short rules for a natural conversational tone; it is intended for dialogue, not for checking supplied text.
The skill does not execute code when activated. Auxiliary Python files under scripts/ are used only when run manually or by GitHub Actions. A release archive contains SKILL.md, README.md, README.en.md, SECURITY.md, SECURITY.en.md, CHANGELOG.md, PERSONA.md, LICENSE, references/ and scripts/.
The README documents use with Claude.ai, Claude Code, API clients that support container.skills, organisation-managed skill libraries, and agent environments supporting the Agent Skills format. For local Claude Code use, it documents placing the skill under ~/.claude/skills/humanizer-ru; copying only SKILL.md omits references/ and reduces the checking depth. The README does not provide a separate runtime command or configuration-key reference for invoking the skill.
Written from the project's own documentation and kept in sync with it. Where the two disagree, the source is authoritative — read the README on GitHub
Same category
zhu1090093659
Skill center for the dsh web GUI: browse all loaded skills grouped by source, enable or disable model invocation, create new skills, and delete into a recoverable trash.
Installed cleanly when we ran it
chuspeeism
现代化可灵活嵌入的任务面板，支持 Codex、DeepSeek Harness
Installed cleanly when we ran it
Alisa0808
Turn one topic into a finished Vox-style paper-collage explainer/ad video — automated end to end on Atlas Cloud + ffmpeg. An agent skill.
Installed cleanly when we ran it
GanyuanRan
Software-engineering method pack for coding agents, with skills for baseline-first planning, systematic debugging, prompt hygiene, verification before completion, and repair/retirement tracking.
Installed cleanly when we ran it
superdesigndev
Design skill for UI and marketing graphics on the Superdesign canvas: reads the repo for context, extracts its design system, then generates and iterates branchable design drafts, flow pages, and reusable components through the Superdesign CLI.
Installed cleanly when we ran it
