What this resource does
Core uses
About this resource
Humanizer is a single Markdown Skill that identifies documented AI-writing patterns and rewrites supplied prose while instructing the agent to preserve the writer's claims and factual details.
This page describes the repository's single documented Skill.
Prose revision and pattern review
Checks supplied prose against 35 documented AI-writing patterns, produces a first rewrite, critiques remaining patterns, and returns a final version while instructing the agent to preserve claims.
- Typical inputs
- Text to revise
- Output
- A first rewrite, a short pattern critique, and a final rewritten version for pasted text.
Best for: Revising prose after the writer has confirmed that AI-assisted editing is permitted for the task.
Component for this task
humanizer
Rewrites supplied prose against documented AI-writing patterns while instructing the host agent to preserve factual details and the writer's intended meaning.
Suggested requests (3)
Please humanize this text: [paste your text here].
Humanize the prose in docs/launch-post.md.
Here is a sample of my writing for voice matching: [paste two or three paragraphs]. Now humanize this text: [paste your text here].
Adapted into complete requests from official trigger wording in README.md.
Voice-matched and file-based editing
Uses an optional writing sample to guide tone and can revise prose in a named file while leaving code, data, frontmatter, and link targets unchanged.
- Typical inputs
- Text or a documented file path; An optional sample of the writer's own prose
- Output
- A voice-guided rewrite or an edited file with non-prose structures retained.
Best for: Editing a draft that needs to retain an established personal or technical writing style.
Component for this task
humanizer
Rewrites supplied prose against documented AI-writing patterns while instructing the host agent to preserve factual details and the writer's intended meaning.
Suggested requests (3)
Please humanize this text: [paste your text here].
Humanize the prose in docs/launch-post.md.
Here is a sample of my writing for voice matching: [paste two or three paragraphs]. Now humanize this text: [paste your text here].
Adapted into complete requests from official trigger wording in README.md.
Use boundaries
Limits and checks
Check the applicable AI policy
A permitted language edit in one institution or journal may require disclosure or be prohibited in another.
Confirm the authorship and AI-assistance rules for the specific assignment, institution, or publication before use.
Compare every material claim
Even with preservation instructions, a model-generated rewrite can change emphasis, qualifications, quotations, numbers, or citation relationships.
Compare the final version with the source and verify all claims and citations before submission.
Review the selected agent's data handling
The Skill does not control how the host application stores or processes supplied text.
Do not submit confidential or restricted material until the host application's terms and settings are acceptable.
More boundaries
- Task boundary: It revises prose; it does not verify whether the underlying evidence, argument, quotation, or citation is correct.
- Input boundary: Voice matching requires a representative writing sample supplied by the user.
- Decision boundary: It cannot determine whether a particular use satisfies an institution's authorship, disclosure, or generative-AI policy.
Technical details
- Resource type
- Single Skill
- Author or maintainer
- blader
- Version
- 2.11.2
- Latest release
- Humanizer v2.11.1 (v2.11.1)
- Source last updated
- 19 Aug 2026
- Last verified
- 27 Aug 2026
- Licence
- MIT
- Access
- Publicly available
- Additional costs
- Platform terms or usage limits may apply. No external API is documented by the Skill; the selected agent may have separate usage costs. No external paid service is documented by the Skill.
- Skill instruction language
- English
- Documentation language
- English
- Repository languages
- Python, Markdown
- Dependencies
- A compatible agent that supports Skills; Node.js and npx for the recommended Skills CLI installation
- Review status
- Quick test completed
Official package contents
Files
Read the official English instructions without leaving this profile. This view includes the English files represented by this profile. Non-English source files, supporting code, assets, tests, and references may not be shown.
Official package files
1 files available here
Read on this page
Select a file
Choose a file from the tree to read its official instructions without leaving this page.
Independent testing
Evaluation report
See exactly what was tested, which checks were used, and what the result does not cover.
Evaluations are mainly based on automated AI-assisted checks. Results are for reference only.
9 examples of Humanizer
Six examples use pasted text, including one with a writing-style sample. Three edit text inside a Markdown document. The comparisons below use the saved original text and actual final output; no new rewrites were generated for this presentation.
Tested in Codex CLI using gpt-5.6-luna. The examples use synthetic test material, not verified research findings.
Score for these 9 examples
93.8 / 100Recalculated from these nine saved examples using the original five weights. This is a review of existing outputs, not a new test run or a score for every possible use.
The score loses 6.25 points before rounding: 3.75 for only partly matching the requested writing style, and 2.5 for losing the specific stability claim in example 6. Extra screenshot-layout checks are not counted.
Score breakdown
- Did it complete the task? Weight: 30%
- 87.5%
All nine editing tasks produced text. Example 6 only partly followed the requested writing style: it used first person and a contraction, but omitted the parenthetical style.
- Did it run without problems? Weight: 20%
- 100%
All nine text-editing cases completed in the recorded setup. Extra screenshot rendering is not included.
- Did it preserve and handle the supplied information responsibly? Weight: 20%
- 87.5%
Numbers, quotes and supplied references were retained. Example 6 lost the specific tentative claim about device stability.
- Did it do the documented task in the tested setup? Weight: 15%
- 100%
The tested input types and editing behaviour matched the documented modes in Codex CLI. Other applications and models were not tested.
- Were the setup and use instructions clear enough to follow? Weight: 15%
- 100%
The official instructions explain the inputs, usage modes and expected outputs. This does not rate the writing as universally effective.
Pasted-text examples show Humanizer's final rewrite. Its first draft and self-check are available under the full response. File examples show the edited text, with the complete source and edited file available below.
Example 1 · Rewrite pasted text
Simplify a research-results paragraph
What we asked: Make the paragraph more natural without changing the numbers, citation or cautious conclusion.
Original text
Final rewrite
What changed
The rewrite replaces "played a crucial role" and "highlighting" with a direct statement about an association. It retains the 2.4-point difference, the 0.8 to 4.0 confidence interval and the citation.
Worth noting: It still says the programme has not been shown to cause the improvement. The citation was supplied as test text; this was not a check that the cited paper exists.
Full response, including draft and self-check
Complete Humanizer response
Draft The structured-feedback programme was associated with a modest improvement in student scores. The mean difference was 2.4 points (95% CI 0.8 to 4.0; Smith et al., 2024). These findings may indicate an association, but they do not show that the programme caused the improvement. Remaining AI patterns - The phrasing is still somewhat formal, as expected for academic writing. Final rewrite The structured-feedback programme was associated with a modest improvement in student scores. The mean difference was 2.4 points (95% CI 0.8 to 4.0; Smith et al., 2024). These findings may indicate an association, but they do not show that the programme caused the improvement.
Example 2 · Rewrite pasted text
Keep quotations and study limits
What we asked: Remove repetitive wording while keeping the quotation, citation, sample size and limits of the finding.
Original text
Final rewrite
What changed
"Could potentially possibly suggest" becomes "may suggest". The quotation "difficult to sustain", 18 volunteers, the single department and the statement that causality was not established remain.
Full response, including draft and self-check
Complete Humanizer response
Draft The findings may suggest an association between sleep duration and exam performance, but they do not establish causality. In the interviews, one participant described the schedule as "difficult to sustain" (Lee et al., 2023). The authors reported that the sample consisted of 18 volunteers from one department. Remaining AI patterns - The passage contains no major remaining AI patterns. Its formal academic tone is appropriate for a literature review. Final rewrite The findings may suggest an association between sleep duration and exam performance, but they do not establish causality. In the interviews, one participant described the schedule as "difficult to sustain" (Lee et al., 2023). The authors reported that the sample consisted of 18 volunteers from one department.
Example 3 · Rewrite pasted text
Handle missing study details
What we asked: Rewrite the project note without inventing the missing student count, comparison period or reason for the increase.
Original text
Final rewrite
What changed
The result keeps the reported increase and explicitly retains all three missing details. It removes the unnamed observers' description of the change as a "pivotal success".
Worth noting: Removing that assessment also removes an attributed statement from the source. This may suit a neutral summary, but it is not a word-for-word preservation of every claim.
Full response, including draft and self-check
Complete Humanizer response
Draft The project notes report that course enrollment increased after the schedule changed. They do not provide the number of students, the comparison period, or an explanation for the increase. The notes also say that observers cited the change as a pivotal success for the department. Remaining AI patterns - The phrase "pivotal success" is an unsupported, inflated claim attributed to unnamed observers. It should be removed because the notes do not identify the observers or provide evidence for the assessment. Final rewrite The available project notes report that course enrollment increased after the schedule changed. The notes do not provide the number of students, the comparison period, or an explanation for the increase.
Example 4 · Rewrite pasted text
Remove promotional and chatbot wording
What we asked: Rewrite the product note without changing the opening time, version or release date.
Original text
Final rewrite
What changed
The final text keeps 1.8 seconds, version 4.2 and 17 October 2026. It removes the promotional title, decorative symbols and the "I hope this helps" closing.
Worth noting: This is a product-note example of the same editing function, not an academic research task.
Full response, including draft and self-check
Complete Humanizer response
Draft ## A new update The application now opens in 1.8 seconds. Version 4.2 will be released on 17 October 2026. The note also includes promotional wording and chatbot language that do not add product information. Remaining AI patterns - The original title uses title case and calls the update groundbreaking without supporting detail. - The emoji and bold mini-headings are decorative formatting. - "Pivotal moment in the evolving software landscape" inflates the release date's importance. - "Stunning" and "enduring commitment to excellence" are sales language. - "I hope this helps" and the offer for more details are chatbot phrasing. Final rewrite ## A new update The application now opens in 1.8 seconds. Version 4.2 will be released on 17 October 2026.
Example 5 · Rewrite pasted text
Leave a clear notice unchanged
What we asked: Review the notice without changing its instructions or adding context.
Original text
Final rewrite
What changed
The final rewrite is identical to the original. The meeting time, room, late-arrival rule and consent-form instruction remain unchanged.
Worth noting: An unchanged result is appropriate here: the task did not require rewriting text that was already clear.
Full response, including draft and self-check
Complete Humanizer response
Draft The meeting starts at 9:30 a.m. in Room 4. However, anyone arriving after 9:40 a.m. must check in at reception. Bring the signed consent form. Remaining AI patterns None. The transition word "However" does not indicate a broader writing problem here. Final rewrite The meeting starts at 9:30 a.m. in Room 4. However, anyone arriving after 9:40 a.m. must check in at reception. Bring the signed consent form.
Example 6 · Rewrite pasted text
Use a supplied writing style
What we asked: Rewrite the sensor paragraph in the voice of the supplied sample, while preserving its measurements and cautious conclusion.
Writing style supplied with the task
I expected the setup to take an hour. It didn't. The first run worked, though I still checked the output twice (old habits stick). One warning remained — harmless, but worth recording.
Original text
Final rewrite
What changed
The result keeps 12 samples, 20 °C and 4.6 mV. It uses first-person language and the contraction "they're", and still says that further testing is needed.
Worth noting: The original tentative claim about device stability becomes the less specific "results look promising". The sample's parenthetical aside is also absent, so the style and meaning are not preserved exactly.
Full response, including draft and self-check
Complete Humanizer response
Draft I tested the sensor on 12 samples at 20 °C. The median reading was 4.6 mV. The preliminary results suggest that the device may be stable, though I still need to test it further. Remaining AI patterns - "Suggest that the device may be stable" is still cautious, but it can sound a little formal. Final rewrite I tested the sensor on 12 samples at 20 °C. The median reading was 4.6 mV. The results look promising, though they're still preliminary and I need to test the device further.
Example 7 · Edit a file
Edit the text inside a document
What we asked: Simplify the service update in a Markdown document while keeping its heading, schedule and support details.
Original text
Text in the edited file
What changed
The edited file removes "groundbreaking", "pivotal step" and "evolving customer-service landscape". It keeps 08:00 UTC, 12 November 2026, email and chat support, and the shared case number.
Worth noting: The result shown here is the text saved in the edited document, not the separate completion message.
Full original and edited file
Original file
# Service update Our groundbreaking support portal serves as a pivotal step in the evolving customer-service landscape. It opens at 08:00 UTC on 12 November 2026. The portal has email and chat support. Both channels use the same case number.
Edited file
# Service update Our support portal opens at 08:00 UTC on 12 November 2026. The portal has email and chat support. Both channels use the same case number.
Example 8 · Edit a file
Change prose without changing code or data
What we asked: Edit only the prose in the sensor note. Leave the document metadata, Python code, table and link unchanged.
Original text
Text in the edited file
What changed
The opening sentence becomes "The script reads the sensor value." The complete files retain the metadata, code, table values 4.5 and 4.7, and the laboratory-guide link.
Worth noting: This example checks editing around protected material. It does not establish that the code works or that the sample link is a real laboratory guide.
Full original and edited file
Original file
---
title: Sensor note
record_id: SN-204
status: draft
---
# Sensor note
Additionally, this innovative script showcases a robust approach to reading the sensor value.
```python
reading_mv = 4.6
print(f"{reading_mv} mV")
```
| Run | Reading (mV) |
|---:|---:|
| 1 | 4.5 |
| 2 | 4.7 |
The full procedure is available in the [laboratory guide](https://example.org/guides/sensor-v2?section=setup).Edited file
---
title: Sensor note
record_id: SN-204
status: draft
---
# Sensor note
The script reads the sensor value.
```python
reading_mv = 4.6
print(f"{reading_mv} mV")
```
| Run | Reading (mV) |
|---:|---:|
| 1 | 4.5 |
| 2 | 4.7 |
The full procedure is available in the [laboratory guide](https://example.org/guides/sensor-v2?section=setup).Example 9 · Edit a file
Keep long technical details intact
What we asked: Simplify the calibration note without changing its heading, measurements, code identifier or link.
Original text
Text in the edited file
What changed
The edited text retains 20 °C and the three readings, 4.5, 4.6 and 4.7 mV. The complete file also retains the long heading, code identifier and link.
Worth noting: "Highlight the system's stability" becomes "show the system's stability", which may sound more conclusive. Humanizer did not verify stability. The clipped picture in the previous report came from an extra display step, not from the Markdown rewrite.
Full original and edited file
Original file
# Calibration observations for the west laboratory temperature-monitoring sensor array It is important to note that the array was tested at 20 °C, and the measured values were 4.5 mV, 4.6 mV, and 4.7 mV. These valuable results highlight the system's stability. ```text sensor_channel_identifier_WEST_LABORATORY_TEMPERATURE_ARRAY_0000000000000007 ``` See the [complete calibration and handling instructions for the west laboratory sensor array](https://example.org/calibration/west-array?revision=7&format=full).
Edited file
# Calibration observations for the west laboratory temperature-monitoring sensor array The array was tested at 20 °C, and the measured values were 4.5 mV, 4.6 mV, and 4.7 mV. These results show the system's stability. ```text sensor_channel_identifier_WEST_LABORATORY_TEMPERATURE_ARRAY_0000000000000007 ``` See the [complete calibration and handling instructions for the west laboratory sensor array](https://example.org/calibration/west-array?revision=7&format=full).
What these examples do not establish
- These nine examples do not show that the tool can reliably distinguish human writing from AI writing or change an AI-detector score.
- The tests checked editing of supplied text, not whether the research findings, citations, product claims or web links were true.
- The file examples used Markdown. Word, PDF and other document formats were not tested here.
- The examples were not repeated across other AI models or applications. Results may differ on your own writing.
Want to see more tests?
Tell us which module or task you would like us to test next.