
What happened
An FA engineer's Claude Desktop skill post-processes pptx output to remove per-character font direct specifications — which pptxgenjs writes as latin/ea/cs overrides via fontFace — while preserving theme references (values starting with +) and the Symbol font.
Why it matters
Without this step, changing the slide master or theme font leaves all text unchanged because direct font specifications take priority over the theme, effectively locking the deck's typography to whatever the generator wrote.
Summaries like this, every morning.
The skill's rules grew out of a specific frustration: generated pptx files look fine when opened, so it is easy to assume they are correct. The font issue is a case in point — the file opens and displays, but a theme-level font change has no effect because every character carries its own font name. The fix is a post-processing pipeline that unzips the pptx, applies a Python regular expression to specific XML elements, and re-zips, deliberately excluding theme references (values beginning with +) and the Symbol font so that special characters like µ do not degrade into plain letters.
The table problem is a second, subtler trap. When XML tables are built or rows and columns are duplicated or deleted, cells can end up after the row's extLst (the row extension information that holds the row ID). PowerPoint then refuses to read the cell contents, leaving only borders visible. A verification environment using LibreOffice renders the same file correctly, and the skill's own validate.py passes, so the defect can escape notice until the file is opened in PowerPoint. The skill therefore treats a successful LibreOffice render as no evidence of PowerPoint compatibility, and requires a scripted count of ordering violations to be reported as zero before output.
The remaining rules address file safety and speaker notes. Existing files are never overwritten; output goes to a new version number, and the code asserts the target does not already exist (assert not os.path.exists(OUT)). Notes are matched to slides by title and body text rather than by slide number, because numbers shift as slides are added or removed during editing. Validation results are reported as command output and counts, extending the series' standing rule that verification not visible in the output is treated as not performed.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
MomongaCode is out under the MIT license, a terminal AI coding agent that works with Ollama, LM Studio, and Ap…

AI agents built a first-person 'spot the anomaly' horror game in Unity, and the separate quality-check agent s…

A developer's AI-operated cloud browser was blocked with HTTP 403 three times in one day when posting YouTube…

In a 2026-06-19 commit (b5af2be32), Shelly added a flow where a natural-language request is parsed by parseAge…

Ziva.sh tested GPT-6 Luna on 10 beginner Godot prompts, with 'Godot 4' specified, no version, and web search;…

From November 22, 2026, Gomumari's target-list AI agent will read a customer's past won-agency list — around 3…
