generate-variant

Pass

Audited by Gen Agent Trust Hub on Feb 23, 2026

Risk Level: SAFECOMMAND_EXECUTIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses local shell commands and npm scripts to perform tasks including JD analysis (npm run analyze:jd), evidence searching (npm run search:evidence), coverage checks (npm run check:coverage), and asset generation (npm run generate:resume, npm run generate:dashboard).- [CREDENTIALS_UNSAFE]: The skill prompts the user to provide a DASHBOARD_PASSWORD as an environment variable directly in the shell command line. This practice can result in the password being stored in plain text within the user's shell history files (e.g., .bash_history).- [PROMPT_INJECTION]: The skill ingests and processes untrusted Job Description (JD) text provided by users.
  • Ingestion points: User-supplied JD text is saved to source-data/jd-{company}.txt and subsequently processed by the analyze:jd script.
  • Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when saving the JD text via shell redirection.
  • Capability inventory: The skill has the capability to execute local scripts, read from the knowledge base, write to the filesystem, and generate PDF resumes.
  • Sanitization: The workflow incorporates quality gates including an evaluation pipeline (npm run eval:variant) to verify metrics and a red-team pipeline (npm run redteam:variant) to scan for secrets, unverified claims, and tone issues.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 23, 2026, 10:56 AM