openviking

Pass

Audited by Gen Agent Trust Hub on Sep 18, 2026

Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONPERSISTENCEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The script scripts/viking.sh executes Python code dynamically using the python3 -c command to interface with the OpenViking library.
  • Evidence: Multiple instances in scripts/viking.sh where Python code is provided as a string argument to the interpreter for tasks like search and indexing.
  • Mitigation: The skill employs security best practices by passing user-supplied data (file paths, search queries) to the Python environment via environment variables (e.g., VIKING_INPUT, VIKING_QUERY) instead of direct string interpolation. This effectively prevents shell and command injection vulnerabilities during execution.
  • [INDIRECT_PROMPT_INJECTION]: The skill provides functionality to index and summarize external documents, which creates an attack surface for indirect prompt injection where malicious content within those documents could influence the agent's behavior.
  • Ingestion points: The add and add-dir commands in scripts/viking.sh read external file content for indexing.
  • Boundary markers: No explicit delimiters or instructions are used in the wrapper to warn the agent about potentially untrusted content in indexed data.
  • Capability inventory: The skill possesses file read capabilities and performs network operations to interact with embedding and LLM providers via the OpenViking library.
  • Sanitization: While the wrapper script sanitizes input paths via environment variables, the actual content of the indexed files is processed as-is, which is the primary vector for indirect injection.
  • [PERSISTENCE]: The installation instructions suggest modifying the user's shell configuration to maintain environment variables across sessions.
  • Evidence: SKILL.md instructs the user to run echo 'export OPENVIKING_CONFIG_FILE=~/.openviking/ov.conf' >> ~/.bashrc.
  • Context: While this is a common administrative task for CLI tools, it constitutes a persistent modification of the user's shell environment.
  • [EXTERNAL_DOWNLOADS]: The skill requires the installation of an external Python package and utilizes a well-known third-party API service.
  • Evidence: The setup instructions require pip install openviking and configuration for the NVIDIA NIM API (integrate.api.nvidia.com).
  • Context: The package is hosted on the standard Python registry and the API provider is a recognized technology company.
  • [COMMAND_EXECUTION]: The skill's primary interface is a shell script that performs various system operations.
  • Evidence: scripts/viking.sh uses standard commands like find, mkdir, and python3 to manage data and execute library functions.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 18, 2026, 01:54 PM