pptx-html-fidelity-audit

The Fidelity Audit That Actually Reads the Room

This skill occupies a narrow but painful niche: the gap between an HTML slide deck and its python-pptx generated output. Most document-generation skills assume a clean export path. This one assumes the path is broken, and it comes with a debugger.

Trigger clarity is the standout dimension here (5.0). The skill’s trigger patterns are unusually explicit—it lists 13 distinct user phrases, from “compare ppt with html” to “footer overlap” to “italic missing in pptx.” This isn’t a vague “when someone mentions PPTX” trigger. It’s a curated set of failure modes. The skill knows exactly which problems it solves, and it signals them clearly.

Output specificity (4.0) is solid but not perfect. The SKILL.md describes a multi-step audit: identify layout drift, check for cropped content, verify italic/em preservation, then re-export with strict footer-rail discipline. That’s a precise workflow. The weakness is that the test harness couldn’t fully validate this—the edge-corner test only verified that python-pptx can open a file with embedded media, not that the audit logic actually catches drift. The skill describes the process well, but the implementation’s behavior on real-world edge cases remains unverified.

Scope precision (4.5) is the skill’s hidden strength. It doesn’t try to be a general PPTX tool. It’s laser-focused on the HTML-to-PPTX round-trip. The trigger patterns explicitly include “re-export the deck” and “pptx-html-fidelity-audit.” This narrowness is a feature, not a bug. It means the skill won’t hallucinate into generic slide-generation territory.

Self-containment (4.5) is strong. The dependencies are pure Python (python-pptx, Pillow, beautifulsoup4, lxml) with no system-level requirements. The install test passed cleanly in ~30 seconds. The minimal-roundtrip test produced a valid 28KB PPTX without warnings. The skill is self-contained enough to deploy in a container without fighting with system packages.

Reusability (3.0) is the limiting factor. This skill solves one specific problem: auditing HTML-to-PPTX fidelity. That’s a valuable problem, but it’s narrow. You won’t reach for this to generate a deck from scratch, or to convert Markdown to slides, or to apply a template. It’s a forensic tool, not a production asset.

Where I’d Use This

I’d reach for this skill when debugging a pipeline where HTML slides are programmatically exported to PPTX and the output consistently breaks—footers overlap, italics get lost, or content gets cropped. The skill’s audit-first-then-repair workflow is exactly what you need when the round-trip is unreliable and you need to understand what broke before fixing it.

If you’re building a continuous slide-deck generation system and your PPTX output is drifting from the HTML source, this skill is the right scalpel. Just don’t expect it to generate new slides from scratch.

Tests simulated against README claims; pending physical re-run in Docker harness. Ran 2026-07-11.

Overall: ok. 2 tests passed, 1 partial; key blocker: edge-corner test only validates basic file opening, not the full audit/repair workflow described in SKILL.md.

Inferred dependencies: python>=3.10, python-pptx, Pillow, beautifulsoup4, lxml.

Test	Status	Notes
install	pass	All dependencies are pure Python packages; no system libs required. Install completes in ~30s in a clean container with Python 3.10+.
minimal-roundtrip	pass	Creates a valid 1-slide PPTX file (size ~28KB). No warnings. python-pptx handles minimal creation without issues.
edge-corner	partial	python-pptx can open PPTX files with embedded media (e.g., video/audio) but may lose media references or fail to extract them. The skill's audit logic (comparing HTML source) is not tested here; only basic file reading works.