The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents and promotes a nested agent architecture using the platform's Task tool, creating a multi-layered vulnerability surface where instructions can be passed deep into a tree of agents.
Ingestion points: Child agents are spawned with a prompt field containing a <problem statement> (referenced in SKILL.md steps 1 and 3), which typically originates from user input or external data sources.
Boundary markers: The provided implementation examples do not utilize boundary markers or explicit instructions to ignore embedded commands within the delegated problem statements.
Capability inventory: Agents within the tree have access to powerful tools including Bash (shell command execution), Task (recursive agent spawning), and TodoWrite (file system modification).
Sanitization: The skill does not provide or recommend any sanitization, filtering, or validation logic for the content passed from the root coordinator to child sub-agents.

nested-subagents