vector-forge

Warn

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides installation instructions that involve executing a remote shell script piped directly into a root shell for the Mull mutation testing framework. This pattern executes unverified remote content with administrative privileges.
  • Evidence: curl -1sLf 'https://dl.cloudsmith.io/public/mull-project/mull-stable/setup.deb.sh' | sudo -E bash in references/mutation-frameworks.md.
  • [PRIVILEGE_ESCALATION]: Multiple steps in the setup instructions require the use of sudo to perform system-level operations, such as creating directories and moving binaries into protected system paths.
  • Evidence: sudo mkdir -p /usr/local/bin /usr/local/lib and sudo cp ... in references/mutation-frameworks.md.
  • [EXTERNAL_DOWNLOADS]: The skill relies on fetching tools, scripts, and binaries from various external sources including GitHub, Cloudsmith, and public package registries.
  • Evidence: gh release download --repo mull-project/mull ... and curl -1sLf ... in references/mutation-frameworks.md.
  • [REMOTE_CODE_EXECUTION]: The installation workflow for the required tools involves downloading and executing remote assets, which presents a potential vector for remote code execution if the sources or transport channels are compromised.
  • Evidence: The installation instructions for Mull and trailmark in SKILL.md and references/mutation-frameworks.md.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external test vectors in JSON format, creating an attack surface where untrusted data could potentially influence agent behavior.
  • Ingestion points: Test vectors are read from JSON files as specified in Phase 2 of the workflow in SKILL.md.
  • Boundary markers: No specific delimiters or instructions are provided to the agent to treat the data as non-authoritative content.
  • Capability inventory: The skill has extensive capabilities to execute shell commands and mutation frameworks across multiple languages.
  • Sanitization: There is no evidence of sanitization or validation of the natural language content within the JSON files.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 17, 2026, 01:17 PM
Security Audit — agent-trust-hub — vector-forge