graphic-case-study
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute local shell scripts and system commands to process generated files.\n
- Evidence: SKILL.md (Step 7) directs the agent to run 'bash /Users/ksd/Desktop/Varnan_skills/frontend-slides/scripts/export-pdf.sh' to generate a PDF document.\n
- Evidence: SKILL.md (Step 7) utilizes the 'open' command to display the generated HTML files to the user.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input that is interpolated into generated HTML files, creating a potential cross-site scripting (XSS) surface when the files are opened in a browser.\n
- Ingestion points: User-provided 'customer_name', 'challenge', 'solution', and 'testimonial' fields in SKILL.md.\n
- Boundary markers: None present to warn the agent about embedded instructions or to delimit the untrusted data.\n
- Capability inventory: Includes file system writes (mkdir, index.html), shell execution (bash), and file opening (open).\n
- Sanitization: No evidence of input escaping or validation was found in the instructions.\n- [DYNAMIC_EXECUTION]: The skill workflow involves generating executable HTML content and subsequently processing it using an external script.\n
- Evidence: SKILL.md describes a process of writing a self-contained HTML file and then running a shell script against it to perform conversion.\n- [EXTERNAL_DOWNLOADS]: The skill relies on external software installation and automated binary downloads for its core functionality.\n
- Evidence: README.md provides 'npx' installation commands for vendor-specific CLI tools.\n
- Evidence: SKILL.md (Step 7) notes that the PDF export functionality triggers automated installation of Playwright, which involves downloading browser binaries.
Audit Metadata