# Cross Functional Deck Assembly > Turns finished material — an analysis, a document, a set of findings — into a deck: one message per slide, the evidence that supports each message named, an explicit cut list, and the rule that a slide with no takeaway line is not finished. Use when existing work has to become slides for a meeting, and when a draft deck is bloated, unsequenced, or ends without an ask. Trigger on 'turn this into a deck', 'build slides from this', 'make a presentation out of this analysis', 'tidy up these slides', 'we present this on Thursday', 'can you produce a pptx'. Not for preparing a governance pack where the decision rights and appendix split matter (use executive-board-pack-preparation), not for a recurring delivery update (use operations-project-status-report), and not for a customer-facing commercial proposal (use sales-proposal-assembly). - Skill: `alihusains/cross-functional-deck-assembly` (Agent Skill) - Install (CLI): `npx skillmds@latest add alihusains/cross-functional-deck-assembly` - Raw SKILL.md: https://api.skillmd.com/api/skills/alihusains/cross-functional-deck-assembly/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Docs & Writing - Author: alihusains (https://skillmd.com/u/alihusains) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/alihusains/cross-functional-deck-assembly --- # Deck assembly from existing material ## Purpose Most bad decks are not badly designed; they are a document paginated. Every slide carries three points, none of them stated, and the audience reconstructs the argument from bullet fragments while the presenter narrates. This skill converts finished material into a sequence where each slide makes exactly one claim, that claim is written on the slide, and the evidence for it is named — and it forces the cut, which is the step that is always skipped. ## Prerequisites - **Inputs:** the underlying material in finished form — not an intent to produce it. If the analysis is not done, stop: a deck built over unfinished analysis hardcodes conclusions that have not been reached. - **Audience and decision:** who is in the room, what they already know, how long you have, and what you need from them at the end. Without the ask, sequencing is arbitrary. If it is not supplied, ask for it before building. - **Runtime, if a `.pptx` is required:** code execution plus a presentation library in the environment — commonly `python-pptx`. Verify by import at step 0 and stop by name if it is unavailable. Offer a slide-by-slide outline in text as the alternative rather than failing at the export step. - **Assets:** where the deck must follow a house template or brand, the template file. Do not approximate a brand from memory. ## Procedure 1. **Verify the prerequisites**, including the library import if a file is to be produced. 2. **Write the ask first, as one sentence.** "We are asking for X by Y." Every other slide either supports it or is cut. A deck whose ask is discovered during drafting is a deck that will be re-sequenced twice. 3. **Write the message list before any slide exists** — the sequence of claims, one line each, that gets the audience from what they know to the ask. Read the list on its own: if it does not hold together as an argument, no amount of slide work will fix it. 4. **One message per slide, written on the slide.** The takeaway line is the headline: a full sentence stating the claim, not a topic label. "Q3 margin" is a label; "Margin fell 3 points because onboarding costs moved into COGS" is a takeaway. **A slide with no takeaway line is not finished** — either write the claim or cut the slide, because a slide whose message you cannot state is a slide the audience cannot receive. 5. **Attach the evidence to each message** and name it: the chart, table, quote or figure, and where it came from. One primary exhibit per slide. If the message needs two exhibits to stand up, it is two messages. If the message has no exhibit, mark it as judgement on the slide rather than letting an assertion pass as a finding. 6. **Cut, using the table below.** This is the step that decides whether the deck works. Expect to remove a third of a first draft. | Slide | Test | If it fails | | --- | --- | --- | | Has a takeaway line | Can you state the claim in one sentence? | Write it, or cut | | Supports the ask | Remove it — is the ask weaker? | Cut if the ask is unchanged | | Earns its exhibit | Does the chart change what anyone believes? | Cut the chart, keep the line | | Method or detail | Does the decision depend on it? | Move to appendix | | Duplicate | Is this claim made elsewhere? | Merge | | Background | Does this audience already know it? | Cut; you are spending their attention | 7. **Sequence for the audience, not the workflow.** Lead with the answer. The order the work was done in is almost never the order it should be heard in; the methodology-first sequence loses the room before the finding arrives. 8. **Move detail to an appendix rather than deleting it.** Anticipated challenges need backup slides; they do not need to be in the flow. Reference them by number from the main body. 9. **Check the numbers on the slides against the source.** Figures get retyped, rounded and re-filtered during deck building, and a screenshot of a filtered dashboard carries the filter with it. Any number that will be quoted back to you goes through `data-analytics-report-qa` first. 10. **Close with the ask slide:** the decision or action requested, the owner, the date, and what happens if no decision is made. A deck that ends on "thank you" ends without a result. 11. **Read the takeaway lines alone, in order.** They must form the argument on their own. This is the whole test — if the story only holds with narration, the deck fails the moment it is forwarded without you. 12. **Time it.** Roughly one message per minute of speaking time, plus discussion. A 20-minute slot with 40 slides means the ask never gets heard. ## Content rules - Never invent a figure, quote, or source to make a slide land. If the supporting number does not exist in the material, the slide states the claim as judgement or it goes. - Do not present a directional estimate with the typography of a precise one; label estimates and ranges as such on the slide. - Charts follow the message, not the other way round: choose the form that shows the claim, and title the chart with the claim. ## Data handling Classification is inherited from the source material and from the audience — decks travel further than any other artefact, so treat client-level, personal or financial-position data as **Confidential** and aggregate before it reaches a slide. Small-cell disclosure matters: a segment containing a handful of entities can re-identify them in an aggregate chart. If personal data or account numbers are supplied for inclusion, flag it and stop. ## Boundaries - A board or steering-committee pack, where the decision rights, options analysis and appendix split are the substance — use `executive-board-pack-preparation`. - A recurring delivery or project update — use `operations-project-status-report`, which owns the RAG and change-since-last discipline. - A customer-facing proposal or pitch — use `sales-proposal-assembly`. - The analysis itself is not finished — this skill does not produce findings; go back to the analysis skill that owns them. - A roadmap communication to a broad internal audience — use `product-roadmap-communication`. ## Hand-offs - **Receives from:** `finance-month-end-variance-analysis`, `data-analytics-report-qa`, `operations-vendor-evaluation` and `cross-functional-spreadsheet-build` (the finished analysis and its exhibits). - **Routes to:** `data-analytics-report-qa` for any figure that will be quoted, `executive-board-pack-preparation` when the audience turns out to be a governance forum, and `marketing-claim-substantiation` where a slide makes a comparative or performance claim that will leave the building.