The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/validate-presenter-mode.mjs uses vm.runInNewContext to evaluate the SPEAKER_NOTES array literal extracted from the generated presentation files. This involves executing code derived from AI-generated output, which is a potential risk if the content is influenced by malicious user data.
[DYNAMIC_EXECUTION]: The script scripts/validate-swiss-deck.mjs utilizes playwright to render and measure generated HTML files. Any JavaScript scripts included in the generated presentation would be executed within the headless browser environment during this process.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted user content (articles, markdown files) and incorporating it into executable HTML/JS files. These files are subsequently processed by validation scripts that use dynamic execution.
Ingestion points: User-provided articles or markdown files (referenced in SKILL.md Step 1).
Boundary markers: The skill uses structured HTML but lacks explicit delimiters or instructions to ignore embedded commands within the user-provided text interpolated into the deck.
Capability inventory: File writing (cp), shell execution (git), dynamic JS evaluation (vm), and browser automation (playwright).
Sanitization: No explicit sanitization or filtering of user-provided content before its inclusion in the generated HTML template was identified.
[EXTERNAL_DOWNLOADS]: The generated output references external resources from well-known services including Google Fonts, Lucide Icons (unpkg.com), MapLibre (unpkg.com), and Motion One (jsDelivr).
[COMMAND_EXECUTION]: The skill guides the agent to perform update checks using git fetch and git pull on its own directory to maintain updates from the upstream repository https://github.com/op7418/guizang-ppt-skill.

guizang-ppt-skill