Skip to content

ValentinNikolaev/agent-skills / agent-plugins/skills/humanizer

humanizer

Verified

Humanize existing prose while preserving voice and meaning, including targeted English pattern edits. Not drafting or translation. Use ai-detector for analysis.

references/sources.md

View on GitHub
# Source decisions

Reviewed on 2026-10-23 using GitHub metadata, pinned source files, licenses, and
package inventories. Upstream descriptions are claims to inspect, not evidence
that detection scores, scientific grounding, or bypass promises work.

## Adapted sources

| Source at inspected commit | License | Selected guidance and changes |
| --- | --- | --- |
| [blader/humanizer](https://github.com/blader/humanizer/blob/8c307f367c91bd974002f10f2be2965d642c13df/SKILL.md) | MIT, copyright 2025 Siqi Chen | Adapted paragraph-level structural editing and source comparison. Removed permissions to add reactions, genre-imposed voice defaults, rigid pattern rankings, and mandatory intermediate drafts. |
| [Aboudjem/humanizer-skill](https://github.com/Aboudjem/humanizer-skill/blob/a58df065367550b6ce40ff3f648335018d8e0589/skills/humanizer/SKILL.md) | MIT, copyright 2026 Adam Boudjemaa | Adapted restraint, protection of meaningful specifics, and the distinction between a style clue and a defect. Removed preset personalities, AI scores, aggressive hedge removal, and automatic profile loading. |
| [Vladimir-Human/humanizer-ru](https://github.com/Vladimir-Human/humanizer-ru/blob/57c72f8aced31bf4c8a98e7c0eae1e3c3e452fb3/skills/humanizer-ru/SKILL.md) | MIT, copyright 2026 Vladimir | Adapted meaning anchors, proportional editing, Russian impersonal constructions, and grammatical-dash preservation. Removed automatic URL/citation cleanup and file editing inferred from a bare path. |

Complete upstream licenses are retained as
[LICENSE.blader-humanizer](../LICENSE.blader-humanizer),
[LICENSE.aboudjem-humanizer](../LICENSE.aboudjem-humanizer), and
[LICENSE.humanizer-ru](../LICENSE.humanizer-ru). Instructions and examples are
substantially rewritten. Attribution does not imply endorsement.

## Compared, not imported

| Source at inspected commit | License finding | Decision |
| --- | --- | --- |
| [harshaneel/humanize](https://github.com/harshaneel/humanize/blob/9c3dec34dc8e0ed9d26de14a869900a714da8f9e/plugins/humanize/skills/humanize/SKILL.md) | MIT | Forced punctuation changes and statistical variation conflict with this skill. Neither the rewrite nor the companion forensic auditor is bundled. Research claims were not validated. The optional user-requested benchmark is a local design, not adapted from this source. |
| [numen-tech/slopornot](https://github.com/numen-tech/slopornot/blob/71bf2ea2862817ed8d87196ddf1290ea37a439d4/skills/agentic-humanizer/SKILL.md) | MIT | Persistent profile management, voice fingerprints, automatic MCP/CLI probing, and unbounded score loops add unwanted state and dependencies. No profiles, integrations, or tools imported. |
| [gencharitaci/humanize-skills](https://github.com/gencharitaci/humanize-skills/blob/a9f9748d067770a5be6f64b3b804f99555193964/skills/humanize-skills/SKILL.md) | CC BY 4.0, not MIT | Broad purpose modes and calibration machinery are unnecessary for the bounded edit. No source text, references, or examples imported. |
| [matsuikentaro1/humanizer_academic](https://github.com/matsuikentaro1/humanizer_academic/blob/468e67f6fd6d93bbfb257d4a685af8db89d0527b/SKILL.md) | Root file states MIT and attribution but does not reproduce the full grant | Medical-paper conventions and a preconfigured author's profile belong to that project. No source material or sentence-length checker imported. |
| [lynote-ai/humanize-text-skill](https://github.com/lynote-ai/humanize-text-skill/blob/9359e17a8d7d54263adb726f4e6c71e0b0940651/SKILL.md) | MIT | Voice targeting and local detector machinery exceed this task. Tree inspection found a root `CLAUDE.md` symlink; it was not followed or imported. No package resources imported. |
| [diaiq/claude-skill-humanizer](https://github.com/diaiq/claude-skill-humanizer/blob/fd111474c8e96ef8dbbf93538089de8b02da7b1c/SKILL.md) | MIT | The skill sends text to a public external API and advertises detector bypass. Neither the upload workflow nor its promises are imported. |
| [borghei/Claude-Skills content-humanizer](https://github.com/borghei/Claude-Skills/blob/2c12c7efaad13a70bdea6e205698a6c4f93a0682/marketing/content-humanizer/SKILL.md) | Root license is Commons Clause + MIT despite skill frontmatter saying MIT | Restrictive terms and voice injection are unsuitable for this adaptation. No text, scripts, or templates imported. |

## Local ownership and routing

The project voice inspected for compatibility keeps its active core in
`social-voice.md`, with craft, language, and task modules. Those files retain
authority and were not copied or modified. Generic humanizing decisions belong
here; the user's narrator, literary references, humour, language ceilings,
punctuation choices, and assessment procedures remain project-owned.

`humanizer` handles an existing passage's naturalness in its current language and
voice, with removing an AI-generated impression as its primary objective.
Voice preservation constrains that edit. The locally authored `ai-detector`
skill owns analysis and external-check mechanics extracted from the former
humanizer benchmark reference. It also replaced `stop-slop` for targeted English
analysis. Targeted edits belong to `humanizer`. The `stop-slop` entry only routed
old invocations to these skills and was removed in v30; no text from it was
carried into either skill.
An ordinary prose edit does not require another skill or an external check.

The UI-copy extension comes from user-authorized maintenance feedback on one
local interface review. Its six gaps inform original, bounded instructions in
[the UI contract](ui-copy.md) and [L1/L2 checks](l1-calques.md), with detector
calibration owned by the sibling skill. No private corpus, glossary, project
paths, scripts or operational instructions are redistributed. Existing licenses
remain unchanged. This is a local in-sample evaluation with prior findings
visible, not independent evidence of reader benefit or detector accuracy.

The user also requested council-based editing decisions. One bounded session
covers edit selection and critique of the exact candidate, with an unchanged
option and reuse of the same advisors. Existing project councils retain their
required stages and authority. Independent review cannot establish actual reader
perception or the author's preference without feedback.

## Intake and security result

The normalized adaptation scores **92/100, ADD** under the repository import
rubric: capability 19/20, reusability 15/15, clarity 14/15, portability 15/15,
safety 14/15, maintenance 9/10, validation readiness 6/10. Remaining limitations
are subjective editorial judgement, continuing overlap with targeted English
editing, and the need for broader real-use voice evaluations. These are skill
intake scores, not text-authorship or human-likeness scores.

Only instructions and complete licenses from the three selected sources are
adapted. Their inspected package trees contain no symlinks or submodules. No
upstream scripts, binaries, installers, hidden configuration, detector data,
credentials, private paths, or dependencies are bundled or executed. Source
links and commands remain reference material. Ordinary use needs only the
requested text and applicable voice guidance. Explicit external checks are
routed to `ai-detector`, not an imported upstream network workflow or a bundled
detector API. Existing third-party notices and licenses remain with humanizer.
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 198kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 194kUpdated Claude CodeCodexVerifiedInstall