The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the design of multi-agent systems where context is passed between nodes via 'handoff' data. This architecture represents a potential surface for indirect prompt injection if the handoff data originates from untrusted sources.
Ingestion points: The handoff property within the edges array of the create-agent-graph tool (SKILL.md).
Capability inventory: Tools for creating and modifying graph topologies and agent configurations (create-agent-graph, update-agent-graph, create-ai-config).
Boundary markers: The instructions do not explicitly suggest delimiters or markers to separate handoff context from the target agent's system instructions.
Sanitization: No specific data validation or sanitization steps are defined for the handoff payloads.
[SAFE]: The skill references official documentation and resources located on the vendor's domain (launchdarkly.com) and utilizes tools specifically designed for the LaunchDarkly environment.

agent-graphs