The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill enables a parent agent to spawn sub-agents and define their initial instructions. This creates a surface for indirect prompt injection where malicious instructions embedded in data processed by the parent could be passed to the sub-agent.
Ingestion points: The parent agent generates a brief for the --prompt parameter of the bb thread spawn command based on its current task or user request.
Boundary markers: Instructions do not include specific delimiters or guardrails to prevent the sub-agent from obeying malicious instructions that might be present in the brief.
Capability inventory: The skill uses bb thread spawn with --permission-mode full and bb thread interactions approve to authorize sub-agent actions.
Sanitization: No sanitization steps are defined for the content injected into the sub-agent's prompt heredoc.
[COMMAND_EXECUTION]: The skill relies on executing various bb CLI commands and a shell script to perform its functions.
Command usage: Executes bb project list, bb provider list, bb thread spawn, and other bb-cli tools to manage sub-agents.
Script execution: Spawns a local script scripts/permission-mode.sh to check the execution environment's security posture.

nagent