mermaid-diagram

Pass

Audited by Gen Agent Trust Hub on Nov 10, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input ($ARGUMENTS) and uses it to generate file names and content which are subsequently passed to shell-based verification tools. \n 1. Ingestion points: The skill parses user requirements from the $ARGUMENTS variable in SKILL.md. \n 2. Boundary markers: No explicit boundary markers or instructions to ignore embedded instructions are present in the processing flow. \n 3. Capability inventory: The skill performs directory creation (mkdir), file writing (.mmd, .md), and shell command execution (mmdc, npx). \n 4. Sanitization: While the skill instructs the agent to use a "kebab-case" naming convention, it lacks explicit sanitization or validation of the generated string before it is interpolated into shell commands. \n- [EXTERNAL_DOWNLOADS]: In Step 4, the skill attempts to run the Mermaid CLI. If the tool is missing from the environment, it uses npx -y @mermaid-js/mermaid-cli@latest to download and execute it from the public NPM registry. This download originates from the maintainers of the Mermaid library (a well-known service) and is a standard utility for the skill's stated purpose. \n- [COMMAND_EXECUTION]: The skill executes shell commands (mmdc or npx) where the input and output file paths are constructed using a string (<diagram-name>) derived from user-controlled requirements. This pattern creates a surface for command injection if the agent fails to strictly enforce the kebab-case naming convention or incorrectly handles shell metacharacters during string interpolation.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 10, 2026, 09:37 AM