The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill utilizes tools like mcp__flow-nexus__agent_spawn and mcp__flow-nexus__sandbox_create to dynamically instantiate specialized agents and isolated environments at runtime. This allows the agent to autonomously extend its execution environment and capabilities based on its current planning objectives.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process complex, untrusted inputs such as complexGoal and constraints to generate execution plans. These plans then drive high-impact operations like agent spawning and task orchestration through mcp__flow-nexus__task_orchestrate, creating a vulnerability where embedded instructions in a goal could potentially manipulate the swarm's behavior.
Ingestion points: Processes complexGoal, constraints, and resourceConstraints inputs in functions like coordinateWithSwarm and decomposeGoal.
Boundary markers: The implementation examples do not specify the use of delimiters or instructions to ignore potential commands embedded within the provided goals.
Capability inventory: The skill has access to agent_spawn, task_orchestrate, sandbox_create, and workflow_create tools for autonomous operation.
Sanitization: No explicit sanitization or validation of external goal text is demonstrated before it influences agent creation and tasking.
[COMMAND_EXECUTION]: The skill makes extensive use of orchestration and execution tools, such as mcp__flow-nexus__task_orchestrate, to perform multi-step actions across different agent instances.

agent-agent