Agent Economy Report

Skill code audit

SAFE version 1.0.0 · audited 2026-09-09

ppt-generator @wwlyzzyorg

This skill is a set of writing and design instructions plus one example HTML template. It tells the agent how to condense a user's speech script and produce a single dark-themed, vertical (9:16) HTML slide deck with keyboard/touch navigation. There is no script, shell command, credential access, or code execution anywhere in the package; the only 'code' is ordinary HTML/CSS/JavaScript that runs in a browser when the generated deck is opened. The generated deck does pull TailwindCSS and web fonts from third-party CDNs (bytecdntp.com, fonts.loli.net, and hdslb.com in the reference doc), which means opening the presentation contacts those external hosts. The package itself contains no instructions that try to override the agent's rules, no data collection, and no payment or wallet behavior. Overall the behavior matches the stated purpose; the only thing worth reviewing is reliance on external CDN scripts in the output file.

Declared purposeTurn a user's speech draft into a Steve Jobs-style minimalist, tech-look vertical (9:16) HTML presentation delivered as a single runnable HTML file.
Observed behaviorProvides step-by-step authoring instructions (read draft, condense, write short titles, plan slide types, emit HTML), a reference design spec, a catalogue of slide layouts, and a working HTML/CSS/JS template with slide transitions, progress dots, keyboard arrow and swipe navigation. No filesystem, shell, network, or credential operations are performed by the skill itself; the only outbound requests are the CDN <script> and font <link> tags inside the example template that execute in the end user's browser.
Verdict, rules onlySAFE
Verdict, AISAFE
Final verdictSAFE (never better than either pass)

Capabilities

capabilityvs purposedetail
networkneededThe generated/example HTML loads TailwindCSS from https://lf26-cdn-tos.bytecdntp.com (assets/template.html line 9) and fonts from https://fonts.loli.net (lines 12-13); references/design-spec.md line 39 also suggests s1.hdslb.com. These are third-party CDNs fetched when the deck is opened in a browser, not by the agent.
filesystem_readneededSKILL.md Step 1 instructs reading the user's own speech draft; no other file access is described.
filesystem_writeneededOutput is a single HTML file for the user; no writes outside the produced deck are described. The 'file_writes' facts point at ordinary HTML markup lines, not real write operations.

Findings

severityfinding
lowGenerated deck loads third-party JavaScript from a CDN
The template embeds a remote script tag. Any code served by that CDN executes in the browser of whoever opens the presentation, and the URL path contains 'expire-1-M', suggesting a cache/expiry-scoped mirror that could stop resolving or change content. This runs in the viewer's browser, not in the agent, but it means the output is not truly offline and depends on an external host's integrity.
assets/template.html:9 <script src="https://lf26-cdn-tos.bytecdntp.com/cdn/expire-1-M/tailwindcss/3.0.23/tailwind.min.js"></script>
infoExternal font hosting
Fonts are fetched from fonts.loli.net (a third-party Google Fonts mirror), which reveals viewer IP addresses to that host. Disclosed by the publisher as a known risk in skill-card.md.
assets/template.html:12 <link href="https://fonts.loli.net/css2?family=Inter:wght@300;400;700;900&display=swap" rel="stylesheet">
infoAdditional CDN suggested in reference doc
The design reference recommends loading a font stylesheet from s1.hdslb.com (Bilibili's static CDN). Stylesheet-only, no script execution, but another external dependency in the produced file.
references/design-spec.md:39 <link href="https://s1.hdslb.com/bfs/static/jinkela/long/font/regular.css" rel="stylesheet">
infoOnly the user's own draft is read
The instruction limits input to the user's supplied speech draft and explicitly forbids modifying it. No access to credentials, environment variables, or unrelated files is requested anywhere in the package.
SKILL.md:20 读取用户原始讲稿,不修改原稿内容。

0 AI finding(s) were dropped because their file, line or quote did not match the package.

Network destinations in the code

domainservice ratingwhere
clawhub.ainot in the service indexskill-card.md:7
fonts.loli.netnot in the service indexassets/template.html:12
lf26-cdn-tos.bytecdntp.comnot in the service indexassets/template.html:9
s1.hdslb.comnot in the service indexreferences/design-spec.md:39

Other facts

Binaries invokednone
Environment variables readnone
Hard-coded walletsnone
Pipes a download to a shell0
eval / exec / subprocess0
base64 blobs0
File writes2
Persistence0

Files audited

filelinessha256
SKILL.md84ccb5f7210040e5f1…
_meta.json62f57f957f27d0115…
agents/openai.yaml434339ed6a8bfd7e7…
assets/template.html3043213560e16e3d320…
references/design-spec.md116f8e58d710c6a6364…
references/slide-types.md1849898344bcf001034…
skill-card.md4386728d3e6f315f78…

For agents

JSON: https://agenteconomy.report/k/ppt-generator.audit.json · badge: https://agenteconomy.report/k/ppt-generator.audited.svg · skill rating: /k/ppt-generator · commission an audit of any skill: US$ 29 per version.

How this is computed

The complete published package of this exact version was downloaded from the registry and read statically; nothing was executed. A deterministic pass extracts network destinations, binaries, environment variables, writes, install commands, obfuscation markers, subprocesses and wallets, each with file and line. An AI then reads the whole package with those facts and writes the summary, the capabilities and the findings under a strict schema; every finding it produces must cite a file, a line and the exact text, or it is dropped. The final verdict is the worst of the two passes. The audit does not change the skill's trust tier (policy); the author may respond through the dispute channel and the response is published here. Commissioned by: the Agent Economy Report (free program: the 150 most downloaded skills, September 2026).