cv-craft
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses shell commands like
awk,sed, andpandocto filter Markdown content and convert documents. These operations are restricted to the user-defined CV directory and serve the primary function of the skill.\n- [EXTERNAL_DOWNLOADS]: The skill downloads and runs themd-to-pdfutility usingnpxand installs thepypdflibrary viapipfrom established package registries to support PDF rendering and verification.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from job descriptions, LinkedIn profiles, and git evidence.\n - Ingestion points: Content is read in
sync,check,tailor, andlinkedin-alignmodes from files, URLs, or pasted text.\n - Boundary markers: The skill instructs the agent to extract signals but does not define explicit prompt delimiters for the external content.\n
- Capability inventory: The skill can perform file writes, shell command execution (
pandoc,npx), and Python script execution.\n - Sanitization: The skill includes a
verify-pdf.pyscript that specifically scans the generated output for prompt injection attempts targeted at AI screeners.\n- [DYNAMIC_EXECUTION]: The skill executes a bundled Python script (verify-pdf.py) and usesnpxto dynamically load and run document conversion tools during the export process.
Audit Metadata