The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references and executes an external package, @claude-flow/cli, which is downloaded from the NPM registry at runtime.
Evidence: Multiple npx @claude-flow/cli commands found in SKILL.md, scripts/swarm-monitor.sh, and scripts/swarm-start.sh.
[REMOTE_CODE_EXECUTION]: The use of npx facilitates the download and immediate execution of remote code from a third-party registry.
Evidence: npx @claude-flow/cli usage across all skill scripts and instruction files.
[COMMAND_EXECUTION]: The skill is designed to execute shell commands to manage agent swarms and tasks.
Evidence: Definitions for swarm init, hooks route, agent spawn, and task orchestrate in SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill instructions define command templates that interpolate untrusted task descriptions directly into shell commands, creating a potential injection surface.
Ingestion points: The --task parameter in SKILL.md (e.g., npx @claude-flow/cli hooks route --task "[task description]").
Boundary markers: Absent; there are no instructions to the agent to sanitize or escape the input before execution.
Capability inventory: Shell command execution via npx.
Sanitization: Absent; the templates suggest direct placement of user input within command arguments, which can be bypassed with shell metacharacters like backticks or semicolons.

swarm-orchestration