The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill indices documents and performs semantic search/summarization, creating a surface where malicious content within processed files could influence agent behavior.
Ingestion points: Local files are read via scripts/viking.py (cmd_add, cmd_add_dir) and the Python API; URLs are supported via the add_resource method as noted in references/python-api.md.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are implemented when processing or summarizing document content.
Capability inventory: The skill has capabilities to read local files, perform network requests to AI providers (NVIDIA API), and output summaries or full content to the agent context.
Sanitization: The skill does not appear to sanitize or filter the content of ingested documents before indexing or retrieval.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the openviking Python package from a public registry (PyPI) to function.
[DATA_EXPOSURE]: The skill is designed to access the local file system for indexing. There is a risk that a user or agent could be instructed to index sensitive paths (e.g., .ssh, .env files) into the OpenViking database, potentially exposing their contents during search or summary operations.
[PERSISTENCE]: The references/setup-guide.md provides instructions to append an environment variable export (OPENVIKING_CONFIG_FILE) to ~/.bashrc. While documented as a legitimate configuration step, this modifies shell profiles to maintain state across sessions.

openviking