PPT Master: turn a short report into a deck people can edit

What the documentation establishes

PPT Master works through an agent with file and shell access and produces editable PowerPoint output. Source ↗

Its default charts are shape-based; supported native chart/table objects require the documented export option. Source ↗

The walkthrough below is an original worked exercise. We have not independently tested this product’s performance.

1. Write the decision before choosing a theme

The audience is a support manager with five minutes to decide whether a pilot deserves two more weeks. The deck should explain what was observed, what remains unknown, and what information an extension would collect. It should not claim the pilot caused an improvement just because two weeks look different.

Prepare three tiny source files in a working folder: pilot.csv with the table below, context.txt describing the observation period and limitations, and brief.txt containing the generation brief. These are exercise filenames we chose, not special product configuration files. Give every figure a source name that can appear in a footnote.

period,tickets,reopened
before,150,18
pilot,120,12

FICTIONAL TRAINING DATA — not company performance
Each period lasts one week. Assume the same counting method.
Ticket mix, staffing, holidays, and seasonal demand were not controlled.
No response-time or customer-satisfaction measurements are available.
Decision: collect two more weeks before deciding on wider adoption.
Do not invent customer quotes, ROI, labor savings, or a causal explanation.

Calculate the two comparisons yourself. Ticket volume falls by 30, or 20% of 150. The reopened share changes from 18/150 = 12% to 12/120 = 10%, a decrease of two percentage points. Those are different units. A slide saying “reopens improved by 2%” would be ambiguous.

2. Set up a workspace with inspectable outputs

For the full-repository route, clone the canonical PPT Master repository and open its root in your agent. The official setup requires Python 3.10+ and dependency installation. The isolated environment below is our suggested wrapper around that documented install step; these commands are for a reader’s local trial and were not run here.

cd ppt-master
python3 -m venv .venv
source .venv/bin/activate
python3 -m pip install -r requirements.txt

Put the exercise files in inputs/. The agent creates an active project under projects/ and reports its path. Use the Default generation flow for this exercise: it confirms the communication and design choices. Quick skips those staged confirmations and durable design decisions.

Keep a short experiment note containing the repository revision, agent host, selected model, and source filenames. Set a personal spending limit in the host or provider before attempting a larger deck. You do not need generated imagery for this exercise, so avoid adding image-generation cost while diagnosing the basic workflow.

3. Give the agent a concrete production brief

Use PPT Master Default generation to make a six-slide, 16:9 decision brief.
Audience: a support manager reading on a laptop and presenting in a meeting.
Sources: inputs/pilot.csv and inputs/context.txt. All data is fictional.
Decision: whether to collect two more weeks of pilot evidence.
First propose the argument and confirm the design choices.
Slide 1: decision and proposed next step.
Slide 2: the two observation periods and what was not controlled.
Slide 3: ticket counts, with a common zero baseline.
Slide 4: reopened shares; label the two-percentage-point difference.
Slide 5: alternative explanations and missing evidence.
Slide 6: owner, next measurement, review date, and stop criteria.
Use editable text and shapes, a light background, and one accent color.
Keep a visible Fictional training data label and source filename on data slides.
No invented claims, stock imagery, customer quotations, or implied causal effects.
Return the active project path and the export path. Describe any unsupported requirement.

Review the proposed argument before production. Slide titles should state a limited conclusion, such as “Ticket volume was lower during the pilot,” rather than a slogan like “Automation transformed support.” If the plan contains six variations of the same chart, assign each slide a distinct job.

Ask for the chart and one dense text slide first if you want an early visual check. Judge both at normal viewing size. This is a practical feedback request, not a promise that every agent host exposes the same preview controls. Resolve factual or layout mistakes before approving the remaining pages.

4. Check the exported argument against a claim ledger

Run one pass that looks only at numbers and units. Check every label, headline, footnote, and speaker note if notes were produced. A correct chart can still sit under an incorrect title. Then do a second pass for the decision: could a reader name the next action without hearing your narration?

For the extension plan, assign an owner in your own copy and choose an actual review date. Record ticket mix and staffing alongside the two existing measures. Set a stopping condition before collecting results, such as suspending the pilot if a critical request is missed. Keep this operational proposal visibly separate from the two observed rows of data.

5. Test what “editable” means in the saved file

Use the exact active-project export path reported by the agent; the documented default is exports/<project-name>_<timestamp>.pptx within that project. Open a copy in the presentation editor your recipient uses.

Default charts are editable shapes. They do not carry a spreadsheet-style chart data table. The FAQ documents --native-charts-and-tables for supported data-backed chart and table exports. Decide which kind of editing the recipient needs before handing the file over.

Do not change only the numeral printed on a bar. The geometry, total, percentage annotation, and explanatory headline must still agree. For recurring monthly reports, data-backed editing may be worth a less customized appearance. For a one-off illustrated argument, shape editing may be enough. Record that tradeoff in the handoff.

6. Make bounded revisions and preserve the evidence

On slide 4 only, the percentage-point annotation overlaps the legend.
Keep the source values, title claim, colors, and all other slides unchanged.
Move or shorten the annotation without shrinking the whole slide.
Report the specific objects changed and export to a new version.
Recheck that 18/150 is 12% and 12/120 is 10%.

If a page looks wrong, compare its authored SVG in svg_output/ with the PowerPoint. A problem already in the SVG points to authoring; a difference appearing only in the deck points toward export or rendering. Default also produces svg_final/ previews; Quick does not.

Keep the input files, final brief, claim ledger, and exported deck together. Add a short README saying which numbers are fictional, which editor you checked, and whether charts are shapes or data-backed objects. If you later replace the sample with real results, redo the claim ledger before reusing the visual design.

Stop when the intended reader can understand the decision and make the expected edit. More animation, more slides, and more decorative images do not repair a weak claim. This exercise has not been generated or opened in PowerPoint by Current; the acceptance procedure is provided so a real trial has an honest definition of success.

Guide history

Revisions preserve the original instructions. A source recheck alone does not create a new revision.

Revision 1 · First guide: original six-slide brief, fictional data, claim ledger, export/editability checks and bounded revisions.

Continue learning