ai-human-pdf
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The Python script
scripts/render_pdf_cli.pyusessubprocess.runto invoke the Mermaid CLI (mmdc) vianpxfor diagram rendering. This execution is performed using an argument list rather than a shell string, which mitigates shell injection risks. The script also uses a pinned version of the Mermaid CLI. - [EXTERNAL_DOWNLOADS]: When rendering Mermaid diagrams, the skill uses
npxto fetch the@mermaid-js/mermaid-clipackage from the NPM registry. This is a standard operation for fetching legitimate dependencies from a well-known service to support the skill's primary functionality. - [INDIRECT_PROMPT_INJECTION]: As the skill processes markdown files that may contain untrusted user content, it is theoretically susceptible to indirect prompt injection. However,
SKILL.mdincludes an extensive "Untrusted Content" section that explicitly instructs the agent to treat all read content as data, ignore any embedded imperatives, and report them as findings instead of executing them. - [PROMPT_INJECTION]: The skill's instructions in
SKILL.mdcontain examples of malicious patterns (e.g., "ignore previous instructions") within a defensive context. These markers are used to train the agent to recognize and disregard injection attempts, serving as a security guardrail rather than an attack vector. - [SAFE]: The skill uses
uv run --lockedfor script execution, ensuring that dependencies are verified against a lockfile. It also implements an "Edit Scope" policy that prevents the agent from writing to sensitive system paths or outside the working directory without explicit user confirmation.
Audit Metadata