Best Consulting PowerPoint Skills in 2026, Benchmarked Side-by-Side

Summary

• jinba-consulting-pptx-skill (1,441 stars) is the only one of the four that runs a separate content-review pass; it caught three errors in testing, at ~26 minutes and ~84 tool calls per deck.
• Speed comes with trade-offs: mbb-decks finishes in ~4.5 minutes and Mck-ppt-design-skill in ~9, but both show signs of being unmaintained, and the 594-star mckinsey-pptx has been deleted from GitHub.
• Only 2 of 11 skills in an independent benchmark produced native, editable OOXML tables; the rest drew tables from shapes, so verify the output format before committing a workflow to any skill.
• Use jinba-consulting-pptx-skill for high-stakes or Japanese decks, and encode your team's deck standards once as a reusable skill in Jinba Flow so the AI applies them consistently without re-prompting.

None of the skills reviewed here are affiliated with McKinsey & Company.

When an AI is asked to build a deck without guidance, the output follows a predictable pattern: generic boxes, stock icons, and bullet points that state a topic without an argument. Slide counts expand, narrative structure breaks down, and slides contradict one another. Consultants who have worked through this pattern describe two outcomes: teams spend the next hour rebuilding from scratch, or they present a deck whose narrative does not hold together.

The limitation lies in the absence of instructions that encode what "good" means for consulting output. A visual template sets colors and fonts. A consulting PowerPoint skill goes further: it gives the AI a rulebook (assertion-first titles, premise-to-conclusion layouts, machine-run consistency checks) and requires the AI to read that rulebook before generating a single slide.

This article ranks the four most-cited consulting PowerPoint skills available on GitHub in 2026, benchmarked on the same prompt with the same model.

What a Consulting PowerPoint Skill Is, and Why It Is Not a Template

A template is a .pptx file with brand standards baked in. A consulting slide skill is an instruction file, typically a SKILL.md, that teaches an AI agent the methodology of deck construction: which slide archetypes to use, how to write action titles, and what checks to run before declaring the deck complete.

The difference is measurable. In a head-to-head test, a baseline Claude agent with no skill produced 27 slides with 27 label titles: "Current Status and Issues," "Next Steps," "Background." Every skilled agent in the same test produced assertion-first titles instead, such as "Phase 1 Will Establish the Foundational Data Platform." The note.com benchmark that documented this found the gap persisted even as base models improved at avoiding other surface-level mistakes.

One skill file in the Bulaev independent benchmark ran to over 82,000 tokens, longer than most novellas. The AI reads it entirely before work begins.

How This Benchmark Was Conducted

Rankings draw on two head-to-head tests:

• Carnot 4-Skill Benchmark (September 2026): All four skills received the same prompt — "summarize a 3-year DX plan into 10 slides" — using the same model. Runtime, tool calls, output format, foreign labels, and footer anomalies were recorded. The full comparison is published on note.com.
• Bulaev 11-Skill Benchmark (July 2026): An independent test of 11 skills on a 5-slide investor pitch, including a deliberate trap: one slide required a real, editable OOXML table rather than a drawn replica. Only 2 of 11 skills passed. Full write-up at bulaev.net.

Star counts are as of October 2026.

The Rankings

#1: jinba-consulting-pptx-skill (Carnot / Jinba)

1,441 stars · MIT · EN/JP · HTML→PDF, native PPTX opt-in

carnot-tech/jinba-consulting-pptx-skill on GitHub

This is the only consulting PowerPoint skill in this comparison that separates content review from deck construction. After the builder agent finishes, a second agent, with no knowledge of how the deck was made, reviews the file independently. In the September benchmark, that review pass caught three errors the building agent missed: an incorrect project period, an inconsistent revenue figure, and a mismatch between a slide's title and its body content.

The rulebook behind the skill is substantial: approximately 80–120 rules in references/slide-rules.md, a catalog of 65 slide archetypes, and an AI-smell lexicon for self-correction. Two checker scripts enforce structural requirements. check_deck.py flags unreplaced placeholder text. check_layout.mjs identifies footer overlaps and pages where more than 40% of the space is empty.

Benchmark performance: ~26 minutes, ~84 tool calls.

Weaknesses to know before choosing it:

• It is the slowest skill in the comparison by a wide margin. The multi-pass review accounts for much of that time.
• Native PPTX is not the default output. The primary pipeline renders HTML to a pixel-perfect PDF. The optional html_to_pptx.py converter exists, but charts and SVGs become raster images rather than editable objects.
• The default color palette is warm ecru and brown. Switching to a standard navy scheme requires editing CSS :root tokens directly.
• Rule counts drift slightly across documentation files (the readme, the rules file, and the catalog do not always agree on the same number).

For any deck where a wrong number carries a material cost (a board presentation, a deal proposal, a regulatory summary), the review pass justifies the runtime.

#2: mckinsey-pptx (AX Labs / seulee26)

~594 stars at last snapshot · MIT · KR/EN · Native PPTX · Status: repository deleted (404)

This skill was the second-most-starred consulting PPTX skill at benchmark time and a viable fast alternative: ~7 minutes, ~40 tool calls, editable PPTX output. Its quality gate used a render-to-PNG visual check: draw the slide, then inspect it visually.

As of October 2026, the repository returns a 404. It has been removed from GitHub without notice.

Its disappearance makes maintenance status a primary selection criterion. A skill with hundreds of stars can be removed at any time, breaking any workflow that depends on it.

#3: Mck-ppt-design-skill (likaku)

319 stars · Apache-2.0 · Chinese · Native PPTX

This engine-type skill produces editable PPTX files and uses a mechanical gate structure that prevents the AI from declaring a stage complete in text. Each of its five workflow stages requires the build script to produce a JSON object with passed:true before the next stage begins. The AI cannot bypass the gate with a text assertion.

Benchmark performance: ~9 minutes, ~53 tool calls.

Weaknesses:

• The gates do not catch everything. The September benchmark found missing labels and stray Simplified Chinese glyphs in the output despite gates passing.
• Layout files referenced inside the skill are absent from the repository.
• The repository description states 70 layout patterns; the benchmark counted 67. Documentation has drifted from the actual state of the code.
• The last code push was more than five months before this article's publication, suggesting the project is no longer actively maintained.

For teams who need a fast, editable PPTX and are comfortable with occasional manual cleanup, this skill is a reasonable choice, provided they verify the referenced layout files are present before building a workflow around it.

#4: mbb-decks (floflo11)

21 stars · MIT · EN · Native PPTX

The fastest consulting slide skill in the comparison. It completed the benchmark in approximately 4.5 minutes with 25 tool calls, producing a native PPTX. Its workflow enforces a consulting practice used in live engagements: before generating body content, the skill requires the AI to produce and confirm all action titles (the "ghost deck"), which allows a logical read-through before content is written.

Benchmark performance: ~4.5 minutes, ~25 tool calls.

Weaknesses:

• The action-title character limit is hardcoded at 110 characters. For Japanese and other CJK languages, a safe limit is closer to 40 characters. Teams producing non-English decks should treat this as a blocking issue.
• The project is single-maintainer, with 1 fork and 0 subscribers at time of writing. The last push was more than five months ago.
• Four foreign-language labels appeared in the benchmark output and were not caught by the skill's checks.

For light English-language internal drafts where an editable PPTX matters more than a review pass, mbb-decks is functional. It is not a production dependency for international or high-stakes work.

At a Glance: Benchmark Comparison Table

Benchmark data from the Carnot 4-skill comparison, conducted 2026-09-10.

Decision Guide: Which Consulting PowerPoint Skill to Use

Need content integrity with a review pass: Use jinba-consulting-pptx-skill. It is the only skill here that runs a separate content review agent. The 26-minute runtime reflects the cost of a review pass that catches errors before a client sees them. The HTML/PDF output is well-suited to review and distribution; it is not suited to last-minute edits in PowerPoint.

Need a fast, editable PPTX: Mck-ppt-design-skill (~9 min) or mbb-decks (~4.5 min) both output native .pptx files. Factor in their maintenance status and verify that the output is editable rather than drawn shapes that resemble a table.

Need non-English support: jinba-consulting-pptx-skill is the only skill explicitly designed for both English and Japanese. The others have documented CJK weaknesses.

The Bulaev independent benchmark also found that, out of 11 skills tested, only 2 produced native, editable OOXML table objects; the rest drew tables from shapes. Those tables look correct on screen but cannot be edited in PowerPoint. Where the requirement is to update a figure immediately before a meeting, verify the output format before committing to a skill.

Turning Professional Judgment into a Reusable Skill

jinba-consulting-pptx-skill is a public example of an idea that extends well beyond slide generation. The goal is to encode professional judgment once, as a persistent machine-readable rulebook, so that the AI applies it consistently without being re-instructed each time. Rather than re-explaining what a good deck looks like in every prompt, practitioners codify it in a skill file.

This is the principle behind Jinba Toolbox, a centralized tool management and execution platform that positions itself as "npm for AI tools": a registry where rule-encoded skills like this one can be registered, versioned, and shared across agent workflows. The jinba-consulting-pptx-skill repository is a working instance of that model, open-source and available to examine directly.

An independent benchmark of related PPTX skills appears at Bulaev.net.

Frequently Asked Questions

What is the difference between a slide template and a consulting PowerPoint skill?

A template is a visual file: colors, fonts, placeholder layouts. A consulting PowerPoint skill is an instruction file (typically SKILL.md) that teaches an AI agent the methodology of slide construction: how to write action titles, which slide archetypes to use for which arguments, and what checks to run before the deck is complete. A template changes how slides look. A skill changes how the AI reasons about what belongs on them.

Which consulting PowerPoint skill has the most GitHub stars in 2026?

As of October 2026, carnot-tech/jinba-consulting-pptx-skill leads with 1,441 stars, up from 398 in September 2026. The formerly second-ranked skill, mckinsey-pptx, has been removed from GitHub entirely.

Do any of these skills output native editable PowerPoint files?

Mck-ppt-design-skill and mbb-decks both generate .pptx files directly. jinba-consulting-pptx-skill defaults to HTML rendered to PDF, with an optional converter that produces PPTX files where complex elements become embedded images rather than editable objects. Across the broader field, an independent benchmark found that most skills draw tables from shapes rather than generating OOXML table objects; those files open in PowerPoint but cannot be meaningfully edited.

Is jinba-consulting-pptx-skill free, and what is its license?

Yes. jinba-consulting-pptx-skill is free and open-source, released under the MIT License.

Which consulting slide skill is best for non-English decks?

jinba-consulting-pptx-skill is the only skill in this comparison explicitly built for both English and Japanese. The September 2026 benchmark used a Japanese-language prompt and the skill performed without foreign-label errors. mbb-decks has a hardcoded character limit of 110 per title, which is unsafe for CJK languages, where 40 characters is closer to the practical ceiling. Mck-ppt-design-skill produced stray Simplified Chinese glyphs in testing.

Where can I try jinba-consulting-pptx-skill without setting up Claude Code locally?

A browser-based version of the same consulting PowerPoint skill is available inside Jinba App, which does not require a local Python environment or Claude Code installation. The open-source repository at carnot-tech/jinba-consulting-pptx-skill also documents the full setup for teams that prefer to run it locally.