The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The ov add-resource command enables the agent to fetch and ingest data from arbitrary external URLs and Git repositories. The documentation includes examples targeting the author's own repositories on GitHub (e.g., github.com/volcengine/OpenViking).
[COMMAND_EXECUTION]: The skill makes extensive use of the ov CLI tool to perform system-level operations, including filesystem management (ov ls, ov rm), resource ingestion, and semantic searching. It also includes ov task watch for managing scheduled background tasks that periodically refresh ingested resources.
[DATA_EXFILTRATION]: Provides functionality to package and write data from the viking:// namespace to local filesystem paths via the ov export and ov backup commands, which could be misused to move sensitive data out of the controlled environment.
[PROMPT_INJECTION]: The core functionality of ingesting untrusted external data creates a surface for indirect prompt injection attacks.
Ingestion points: Untrusted data enters the agent context through ov add-resource (URLs, Git repos, local files), ov import (.ovpack archives), and ov write (manual content creation).
Boundary markers: The provided documentation does not mention the use of delimiters or instructions for the agent to ignore embedded commands within ingested content.
Capability inventory: The skill grants the agent capabilities to modify the environment, including file writes (ov write), deletions (ov rm), and directory management (ov mkdir).
Sanitization: There is no evidence of content sanitization, validation, or filtering for external resources before they are added to the semantic context.

ov-resources