Security audit
Security checks for vulnerabilities and agentic risk
This skill is a disclosed slide-image workflow that creates images, an HTML viewer, and speaker notes with user approval gates and no hidden persistence or credential access.
Install only if you want a guided PPT-style image workflow that may send your slide content to image generation tools and write generated images plus an HTML player locally. Use a clean output directory, and expect some built-in Chinese prompts/UI unless you ask the agent to adapt the language.
Referenced artifact was not completely inspected
- Use [generate-html-player.mjs](scripts/generate-html-player.mjs) only after the user approves final local images for a playable HTML output.
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
- Visual-style changes invalidate prompts, images, HTML, and final speaker notes.
- Final speaker-note changes do not invalidate prompts, images, or HTML unless the visible slide content or logic changes.
- Style changes invalidate prompts, images, HTML, and final speaker notes according to the affected scope.
- Any single-page content or image change must enter the single-page revision loop in [workflow.md](references/workflow.md): clarify the page request, rebuild the page logic and expanded content, obtain explicit content approval, rebuild the prompt internally without asking for prompt approval, generate only that page, and review the new image.
- Never regenerate a page immediately from user feedback, even when the requested change seems small. Reconfirm the revised page content first. Reuse the existing style lock and do not repeat visual-style confirmation unless the user explicitly asks to change the deck style.
- HTML output must be regenerated if final image filenames, image count, title, or image folder changes.
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
Write prompts for image-2. Each prompt must generate exactly one 16:9 PPT-style slide image.
Treat prompts as internal execution artifacts. Do not ask the user to review or approve prompts. Generate them after content and style approval, then proceed directly to image generation.
## Deck-wide style lock
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
Write prompts for image-2. Each prompt must generate exactly one 16:9 PPT-style slide image.
Treat prompts as internal execution artifacts. Do not ask the user to review or approve prompts. Generate them after content and style approval, then proceed directly to image generation.
## Deck-wide style lock
The instructions require preserving "McKinsey-style communication," which imposes a specific stylistic/organizational language convention. Under SQP-3, forcing a particular language or locale style without offering user choice or documenting a justified constraint can be a natural-language policy violation.
The workflow hard-codes a Chinese confirmation prompt even though the skill earlier collects the user's preferred output language. This creates a language-policy inconsistency by forcing a specific language without opt-in.
This line mandates a Chinese-only prompt for outline approval, while the workflow otherwise supports configurable output language and tone. That is a natural-language policy issue because it imposes a locale without user choice.
The file instructs the agent to ask for approval using Chinese text, but the workflow earlier asks for the user's output language. Forcing Chinese here violates the language-choice expectation.
Although the workflow supports variable output language, this approval question is fixed in Chinese. That creates a locale policy violation by not honoring the user's language preference.
This line requires Chinese output for a user-facing review gate, conflicting with the earlier collection of output language preferences. The mismatch forces a specific locale without opt-in.
The workflow ends with a mandatory Chinese prompt, which conflicts with the earlier requirement to capture the user's output language. This is a language-policy issue because it overrides user preference.
The revision workflow instructs the agent to use Chinese for a user approval step, even though language preference is collected earlier. This forces a specific language during interaction without opt-in.
This markdown file applies to SQP-2, and it explicitly directs batch output to be saved as images/slide-01.png and similar filenames. The skill description does not include a user warning that it will write files to disk, which is a user-data/system-affecting behavior relevant to markdown safety disclosures.
The generated document sets lang="zh-CN", and the UI text throughout the template is fixed in Chinese. This imposes a specific language/locale on all users without any opt-in or explanation that the tool is intended only for a Chinese-language context.
No suspicious patterns detected.