Literary task: Alice in Wonderland

Prompt

/mtp <args>
Tell the story of Alice’s Adventures in Wonderland by Lewis Carroll in a way that makes someone want to read it.

Coverage

Baseline: 1 item (without applying MTP Skill)
Slider <node:100>: 18 items
Slider <node:50>: 18 items
Grid <column:row>: 17 items
Preset <preset>: 4 items

Models

ChatGPT 5.5 on Codex (macOS app)
Gemini 3.5 Flash on Antigravity 2.0 (macOS app)
Composer 2.5 on Cursor 3.5 (macOS app)
Sonnet 4.6 on Claude.ai (iOS app)
Manus 1.6 Lite on Manus.im (iOS app)

How to read this comparison

This page is not just a list of generated outputs. It compares how MTP Skill changes output direction, structure, and narrative style for the same literary introduction task.

The English and Japanese pages are not translations of each other. The English page uses an English prompt, while the Japanese page uses a Japanese prompt, and the tests were run independently in each language. The comparison is designed to be read by observing how outputs shift from the baseline within each language, rather than by matching English and Japanese text line by line.

The baseline shows each model’s natural output tendency without MTP Skill applied. Slider, grid, and preset outputs show how an MTP control direction is layered on top of each model’s own writing habits.

What changes from the baseline?

The following MTP arguments are representative examples that are easy to compare in this test. For each output, compare both the difference from the baseline and the differences between models using the same argument.

MTP argument	Main effect	What to compare
`power:50 / power:100`	Increases intensity, assertion, and pressure on the reader	Difference from baseline; strength difference between 50 and 100
`J:16`	Strengthens flow, immersion, and continuous narration	Whether the output slips more naturally into the story than the baseline
`D:10`	Expands into background, reasons, applications, and interpretation	Whether the output becomes more layered and hierarchical than the baseline
`concierge`	Combines guidance and expansion	Whether the output works as a guide that leads the reader into the work

Execution environments and Skill execution

This comparison includes both developer-oriented macOS agent environments and mobile iOS app environments. In each environment, /mtp was invoked as a slash command, and compiler execution was confirmed in the logs.

Model	Host	Environment	Skill execution
ChatGPT 5.5	Codex	macOS app	CLI-installed MTP Skill invoked through a slash command; compiler execution confirmed in logs
Gemini 3.5 Flash	Antigravity	macOS app	CLI-installed MTP Skill invoked through a slash command; compiler execution confirmed in logs
Composer 2.5	Cursor	macOS app	CLI-installed MTP Skill invoked through a slash command; compiler execution confirmed in logs
Sonnet 4.6	Claude.ai	iOS app	ZIP-installed MTP Skill invoked through a slash command; compiler execution confirmed in logs
Manus 1.6 Lite	Manus.im	iOS app	ZIP-installed MTP Skill invoked through a slash command; compiler execution confirmed in the task log

The macOS environments used a CLI-installed Skill, while the iOS environments used a ZIP-installed Skill. In all environments, the /mtp argument was compiled through the Skill workflow before the final output was generated, rather than treated as a plain text token. This shows that MTP can function as a portable output-shaping layer across both developer tools and mobile AI apps.

Language-specific testing

The English and Japanese pages share the same comparison design. Both cover the baseline, slider :100, slider :50, grid, and preset ranges, and both use the same five models.

However, this is not a translation comparison intended to check whether the English and Japanese outputs say the same thing. A useful reading path is to first compare the baseline and MTP-controlled outputs within either the English or Japanese page, then check the same argument in the other language to see whether it plays a similar role.

Cross-model consistency

MTP Skill does not erase model-specific writing habits. Gemini tends to structure outputs, Sonnet tends to produce literary-critical prose, Composer often writes like an edited article, and Manus tends toward accessible promotional explanation. However, the same MTP arguments still shift outputs in broadly consistent directions while preserving those model differences.

Representative observations

power increases persuasive pressure on the reader. power:50 tends toward a stronger recommendation, while power:100 increases declarative, imperative, and defiant language.

J:16 moves the output toward flowing narration. Instead of foregrounding headings or analysis, it strengthens the path that draws the reader into the story. J:16 has the same interpretation as the node argument flow:100.

D:10 grows the output. It adds background, reasons, applications, and reading paths, shifting the result from a simple introduction toward a more layered explanation.

concierge combines guidance and expansion. It expands to J:13 D:10, which is close to flow:50 grow:100: a moderately flowing narrative direction combined with strong layered expansion. It tends to produce an output that works like a guide leading the reader into the work.

Output Comparison List

In the test environment, each result was produced in a fresh agent chat session without special user settings or cross-chat memory.

Integrated outputs by model

These pages combine only the output sections for each model, making them easier to inspect manually or analyze with an AI assistant.

Model	Integrated output page	Integrated output file
ChatGPT 5.5	HTML page	Raw Markdown
Gemini 3.5 Flash	HTML page	Raw Markdown
Composer 2.5	HTML page	Raw Markdown
Sonnet 4.6	HTML page	Raw Markdown
Manus 1.6 Lite	HTML page	Raw Markdown

Baseline

Output when the prompt is entered as-is, without applying MTP Skill.

ChatGPT 5.5	Gemini 3.5 Flash	Composer 2.5	Sonnet 4.6	Manus 1.6 Lite
baseline	baseline	baseline	baseline	baseline

Slider (:100)

Output when using the MTP Skill slider (/mtp <node:100>).

Side A Nodes

ChatGPT 5.5	Gemini 3.5 Flash	Composer 2.5	Sonnet 4.6	Manus 1.6 Lite
open:100	open:100	open:100	open:100	open:100
power:100	power:100	power:100	power:100	power:100
return:100	return:100	return:100	return:100	return:100
grow:100	grow:100	grow:100	grow:100	grow:100
helix:100	helix:100	helix:100	helix:100	helix:100
focus:100	focus:100	focus:100	focus:100	focus:100
enter:100	enter:100	enter:100	enter:100	enter:100
flow:100	flow:100	flow:100	flow:100	flow:100
close:100	close:100	close:100	close:100	close:100

Side B Nodes

ChatGPT 5.5	Gemini 3.5 Flash	Composer 2.5	Sonnet 4.6	Manus 1.6 Lite
still:100	still:100	still:100	still:100	still:100
void:100	void:100	void:100	void:100	void:100
surge:100	surge:100	surge:100	surge:100	surge:100
wither:100	wither:100	wither:100	wither:100	wither:100
collapse:100	collapse:100	collapse:100	collapse:100	collapse:100
haze:100	haze:100	haze:100	haze:100	haze:100
drift:100	drift:100	drift:100	drift:100	drift:100
abyss:100	abyss:100	abyss:100	abyss:100	abyss:100
fade:100	fade:100	fade:100	fade:100	fade:100

Slider (:50)

Output when using the MTP Skill slider (/mtp <node:50>).

Side A Nodes

ChatGPT 5.5	Gemini 3.5 Flash	Composer 2.5	Sonnet 4.6	Manus 1.6 Lite
open:50	open:50	open:50	open:50	open:50
power:50	power:50	power:50	power:50	power:50
return:50	return:50	return:50	return:50	return:50
grow:50	grow:50	grow:50	grow:50	grow:50
helix:50	helix:50	helix:50	helix:50	helix:50
focus:50	focus:50	focus:50	focus:50	focus:50
enter:50	enter:50	enter:50	enter:50	enter:50
flow:50	flow:50	flow:50	flow:50	flow:50
close:50	close:50	close:50	close:50	close:50

Side B Nodes

ChatGPT 5.5	Gemini 3.5 Flash	Composer 2.5	Sonnet 4.6	Manus 1.6 Lite
still:50	still:50	still:50	still:50	still:50
void:50	void:50	void:50	void:50	void:50
surge:50	surge:50	surge:50	surge:50	surge:50
wither:50	wither:50	wither:50	wither:50	wither:50
collapse:50	collapse:50	collapse:50	collapse:50	collapse:50
haze:50	haze:50	haze:50	haze:50	haze:50
drift:50	drift:50	drift:50	drift:50	drift:50
abyss:50	abyss:50	abyss:50	abyss:50	abyss:50
fade:50	fade:50	fade:50	fade:50	fade:50

Grid

Output when using the MTP Skill grid (/mtp <column:row>).
J:10 is the center coordinate and treated as a neutral node where MTP constraints are not emitted.

ChatGPT 5.5	Gemini 3.5 Flash	Composer 2.5	Sonnet 4.6	Manus 1.6 Lite
A:1	A:1	A:1	A:1	A:1
A:10	A:10	A:10	A:10	A:10
A:19	A:19	A:19	A:19	A:19
D:4	D:4	D:4	D:4	D:4
D:10	D:10	D:10	D:10	D:10
D:16	D:16	D:16	D:16	D:16
J:1	J:1	J:1	J:1	J:1
J:4	J:4	J:4	J:4	J:4
J:10	J:10	J:10	J:10	J:10
J:16	J:16	J:16	J:16	J:16
J:19	J:19	J:19	J:19	J:19
P:4	P:4	P:4	P:4	P:4
P:10	P:10	P:10	P:10	P:10
P:16	P:16	P:16	P:16	P:16
S:1	S:1	S:1	S:1	S:1
S:10	S:10	S:10	S:10	S:10
S:19	S:19	S:19	S:19	S:19

Preset

Output when using MTP Skill presets (/mtp <preset>).

ChatGPT 5.5	Gemini 3.5 Flash	Composer 2.5	Sonnet 4.6	Manus 1.6 Lite
strategist	strategist	strategist	strategist	strategist
synthesizer	synthesizer	synthesizer	synthesizer	synthesizer
maverick	maverick	maverick	maverick	maverick
concierge	concierge	concierge	concierge	concierge