jfrog-ai-catalog-skills

Pass

Audited by Gen Agent Trust Hub on Jul 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes various commands using the jf CLI and shell utilities like rm, test, and jq for file system management and API interaction.\n- [EXTERNAL_DOWNLOADS]: The skill downloads agent skill archives from JFrog Artifactory and fetches the @jfrog/agent-guard tool from a remote npm registry.\n- [REMOTE_CODE_EXECUTION]: The skill runs the @jfrog/agent-guard utility using npx directly from a specified registry, which is a form of remote code execution used for management tasks.\n- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing external skill names and descriptions.\n
  • Ingestion points: External data is ingested through the catalog discovery commands (--list-skills, jf skills list).\n
  • Boundary markers: The instructions mandate using Markdown tables for presenting external results to the user.\n
  • Capability inventory: The skill environment includes shell execution (jf, npx), file system access (rm, test), and network access via the JFrog API.\n
  • Sanitization: The skill uses structured output templates and requires explicit user consent for all mutating operations (installing, removing, or publishing) to mitigate the risk of malicious embedded instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 22, 2026, 02:49 AM
Security Audit — agent-trust-hub — jfrog-ai-catalog-skills