The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides tools like add_resource to import external data, including web pages and Git repositories, into the agent's long-term memory. This creates a surface where an attacker could embed malicious instructions in those external sources, which may be followed by the agent when that context is later retrieved and processed.
Ingestion points: The add_resource tool (accepts URLs and Git repos), the remember tool, and the automatic session capture mechanism described in SKILL.md.
Boundary markers: The instructions suggest using an <openviking-context> tag for injected context and note that live conversation should take precedence over recalled memories, which acts as a partial mitigation.
Capability inventory: Capabilities include repository indexing, file system reading via read and grep, and command-line interactions through the ov CLI.
Sanitization: The skill documentation does not specify any sanitization, filtering, or validation performed on the ingested content before it is stored or recalled.
[EXTERNAL_DOWNLOADS]: The skill contains references to external documentation and setup guides on the openviking.ai domain. These are used to provide instructions for the required CLI tools.
Evidence: Links to docs.openviking.ai in the 'Beyond the MCP tools' section of SKILL.md.

openviking-memory