The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of an external package from PyPI: pip install clawteam. This introduces a third-party dependency into the environment.\n- [COMMAND_EXECUTION]: The clawteam spawn command is used to launch sub-processes (defaulting to the claude CLI agent). This allows the skill to execute and manage external programs dynamically.\n- [REMOTE_CODE_EXECUTION]: The skill documentation explicitly instructs users and agents to use the --dangerously-skip-permissions flag during the spawn process. This instruction is designed to bypass standard tool-use confirmation and safety guardrails for sub-agents, allowing them to perform potentially dangerous operations autonomously without human oversight.\n- [PROMPT_INJECTION]: The skill implements an inter-agent messaging system (clawteam inbox) which acts as a vector for indirect prompt injection (Category 8). \n * Ingestion points: Data enters the agent context via clawteam inbox receive (found in SKILL.md and references/workflows.md). \n * Boundary markers: None identified; messages from other agents are processed as plain-text instructions. \n * Capability inventory: The skill allows spawning sub-processes, writing files to ~/.clawteam/, and executing CLI tools. \n * Sanitization: No evidence of input validation or escaping for incoming messages.

ClawTeam Multi-Agent Coordination