The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to fetch external advisories and specifications using an untrusted third-party proxy service (curl -sL "https://defuddle.md/<url>"). While the skill provides usage guidelines and OPSEC warnings, using this service sends all requested URLs to a third-party server.
[PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted external content into the agent's context, creating a surface for indirect prompt injection.
Ingestion points: The agent is instructed to fetch external documentation via the defuddle.md proxy (SKILL.md).
Boundary markers: The skill lacks explicit prompt delimiters or instructions to treat the fetched markdown as untrusted data, although it does warn against using the service for adversary infrastructure.
Capability inventory: The skill utilizes kubectl for cluster administration and curl for network requests, providing significant capabilities that could be targeted by injected instructions.
Sanitization: No sanitization or validation of the fetched content is described beyond the proxy's boilerplate removal.
[COMMAND_EXECUTION]: The skill contains several shell commands using kubectl, jq, and curl. These allow the agent to interact with Kubernetes APIs and external networks. While these are standard administrative tools, they constitute a capability set that requires careful use when combined with external data ingestion.

defending-kubernetes