The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to use 'npx' to download and execute the 'claude-flow' package from the NPM registry. While NPM is a well-known service, the package is executed without specific version pinning, relying on the latest available version at runtime.
[COMMAND_EXECUTION]: The skill provides various commands for initializing swarm topologies, spawning queen/worker agents, and monitoring consensus status through the 'claude-flow' utility.
[INDIRECT_PROMPT_INJECTION]: The implementation of Byzantine fault-tolerant consensus mechanisms involves processing data and state from multiple distributed agents. This architectural pattern creates a surface where one compromised or malicious agent node could attempt to inject instructions into the collective decision-making process. (1) Ingestion points: Swarm coordination and consensus strategies defined in SKILL.md. (2) Capability inventory: Shell command execution and multi-agent coordination via 'npx claude-flow'. (3) Boundary markers: None explicitly defined in the skill instructions. (4) Sanitization: No sanitization or validation protocols for inter-agent communication are specified in the instructions.

hive-mind