MBB Decks
Generate real consulting-grade .pptx files from Claude, with action-title storylines, MECE bullets, and company logos as bullet markers.
Data flows are well disclosed: logo.dev downloads, HEAD probes of source URLs, the shared public token and the LOGODEV_TOKEN override are all documented; the build script references local files with no sign of overreach or concealment. Deducted for: automatic outbound network requests at render time without user confirmation, shared token quota, no stated rollback/isolation, and unverified publisher identity.
Internally consistent: dependency self-check, failure message script for missing python-pptx, slide-count validation warning, and controlled degradation (logo 202/4xx falling back to plain bullets) are all specified. Deducted for: static review only, no test evidence, scripts/build_deck.py absent from the evidence set, and the non-4xx silent-keep behavior unverifiable from source shown.
Trigger conditions are clear (MBB-style decks, board prereads, DD decks, etc.), boundaries like the ghost-deck flow and 15-slide cap are explicit, and limitations (only clustered column charts render natively) are declared. Deducted for: no statement on logo.dev / paywalled-source reachability from mainland China, no Chinese-language support, and undefined offline behavior for logo fetching.
SKILL.md is well layered: workflow, structure, visual system, JSON schema, common-mistakes table, red-flag list, MIT license, and author background, plus two worked example directories. Deducted for: no version number, no changelog, no known-limitations or maintenance/update-path statement, and the shared-token exhaustion risk is not disclosed as a limitation.
Goals are concrete (real .pptx, action-title storylines, MECE, chart-plus-commentary co-location) and README claims committed comparison decks and examples, giving a clear value proposition over manual deck building. Deducted for: static review cannot confirm outputs are directly usable; the script and example files are not in the evidence set; marketing-style claims ('not slop') lack executed corroboration.
There are partial auditable traces (two example input JSONs, comparison images, MIT license) and an invitation to inspect the committed decks. Deducted for: this evidence set contains only SKILL.md/README/LICENSE; the build script and examples are missing; no third-party execution evidence (CI, tests); comparison images are author-supplied and coverage is thin.
- The skill makes automatic outbound network calls to logo.dev and performs HEAD probes of source URLs at render time; offline or restricted networks degrade logos and make link validation non-deterministic. Mainland-China users may lose the logo bullet feature if logo.dev is unreachable.
- A shared public logo.dev token ships with the skill; it is a public quota that may be exhausted or rate-limited. Register your own token and export LOGODEV_TOKEN.
- This is a static review; no scripts were executed. scripts/build_deck.py and the example files were not in the evidence set, so output quality and reproducibility are independently unverified.
- No version number or changelog exists, so behavior may shift between pulls without traceability.
What does this skill do, and when should you use it?
MBB Decks is a Claude Agent Skill that produces McKinsey/Bain/BCG-style PowerPoint presentations as real .pptx files, not Markdown outlines. It codifies consulting house-style rules: action titles that read as a coherent storyline, a six-section deck structure, 3-5 MECE bullets per slide, and a navy-and-white editorial visual system. The rules come from three years of the author drafting decks alongside ex-MBB executives, and the author (Iris Meng) spent three years at EY NYC advising $40B+ M&A deals. It fits anyone who needs boardroom-ready output for executive audiences.
The skill takes a topic or analysis and runs a four-step workflow: (1) it produces a 'ghost deck' containing only each slide's action title and verifies they read as one coherent argument; (2) it shows the ghost deck and waits for user confirmation; (3) it expands into bullets, charts, footnotes, and sources in a JSON spec; (4) it calls scripts/build_deck.py (python-pptx based) to render a .pptx. At render time it converts markdown-style [text](url) sources into clickable hyperlinks, HEAD-checks each URL and silently drops 4xx links, auto-downloads company logos from Logo.dev as bullet markers, and enforces a 15-slide hard cap on the main deck.
- A consultant or strategy team delivering market entry, due diligence, or recommendation decks to client executives at MBB formatting standards
- An analyst who has completed analysis and needs to turn it into a board pre-read or executive readout quickly
- A finance or banking team preparing summary slides where every numeric claim needs a clickable, attributed source
- Anyone building an industry landscape report that names companies and wants real brand logos as bullet markers
- A user who wants an editable PowerPoint file to refine further, not just a text outline
What are this skill's strengths and limitations?
- Outputs a real, editable .pptx that opens in PowerPoint or Keynote for further refinement
- Ghost-deck-first workflow structurally guarantees storyline quality before any body content is written
- Sources render as clickable hyperlinks, verified at render time with dead 4xx links silently dropped
- Real company logos auto-fetched and cached, replacing abstract icons for entity-focused bullets
- Highly specific, enforceable rules: title word counts, bullet limits, a claim-to-chart-type mapping table, and an authoritative source whitelist
- Full example decks and side-by-side comparison images in the repo let you verify quality before adopting
- Chart rendering currently only supports clustered column charts cleanly; other chart types fall back to bullets and require manual charting or script extension
- Requires Python 3.9+ and python-pptx; users without Python must install them first
- Logo feature depends on Logo.dev; the shared public token may need replacing with your own LOGODEV_TOKEN for private quota
- The comparison images are author-supplied; there is no independent evaluation and no publicly documented test suite
- The 15-slide hard cap and fixed visual system are deliberate constraints that limit fully custom styling
How do you install this skill?
Run npx skills add floflo11/mbb-decks and pick your agent when prompted (Claude Code, Codex, Cursor, Gemini, OpenCode, and others are supported). On first run the skill self-checks for Python 3.9+ and python-pptx and tells you the one install command if anything is missing.
How do you use this skill?
After installing, prompt your agent: 'Build me an MBB-style deck on [your topic]. Use the mbb-decks skill.' The skill first shows a ghost deck of action titles for storyline confirmation, then expands to full content and renders a .pptx. Charts, two-column layouts, and logo-marked bullets are driven by a JSON spec; complete worked examples are in examples/market-entry and examples/data-center-landscape.
How does this skill compare with similar options?
The README's side-by-side comparison shows that default Claude output is loose bullet lists, while this skill produces storyline-driven decks following consulting house conventions - the clear contrast is with simply asking an AI to 'make slides.' For financial-model output, the author has announced a companion Excel skill (three-statement models, LBO, valuation builds) still on the roadmap.