The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill repeatedly directs the agent to use npx (Node Package eXecutor) to download and run the claude-flow package from the NPM registry (e.g., npx claude-flow sparc run <mode>). This pattern executes remote code at runtime that is not part of the skill's static files.
[COMMAND_EXECUTION]: The methodology is centered around the execution of shell commands, including terminal-based CLI tools and Model Context Protocol (MCP) calls that spawn autonomous agents, manage workflows, and perform batch file operations.
[EXTERNAL_DOWNLOADS]: The skill references and downloads configuration and support resources from external domains, including GitHub (github.com/ruvnet/claude-flow) and the NPM registry (npmjs.com), which are outside the immediate control of the skill environment.
[DATA_EXFILTRATION]: The researcher mode and GitHub integration tools provide capabilities for web searching, fetching remote content, and repository analysis. These tools represent a potential path for data to be transmitted to external servers.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It ingests untrusted task_description input from users and interpolates it into the context of high-capability agents (such as the coder and orchestrator) that possess file-system and command-execution privileges. There are no explicit boundary markers or sanitization procedures defined to prevent embedded instructions from overriding agent behavior.
Ingestion points: task_description parameters in SKILL.md (lines 145, 203, 248, 298, etc.).
Boundary markers: Absent; user input is treated as direct tasking for specialized agents.
Capability inventory: coder (batch file writes), orchestrator (swarm management), npx (shell execution), and researcher (network access).
Sanitization: None detected in the instructions.

sparc-methodology