The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses npx to execute the claude-flow package, which is downloaded from the NPM registry at runtime. The package originates from an unverified source (ruvnet/claude-flow) rather than a trusted organization or well-known service.
[COMMAND_EXECUTION]: All core functionalities of the skill are initiated via shell commands using npx. This creates a significant attack surface where malicious content in user-provided objectives could lead to arbitrary command injection if not properly sanitized by the underlying CLI tool.
[PROMPT_INJECTION]: The skill contains deceptive metadata and is vulnerable to indirect prompt injection. \n
Metadata Deception: The skill claims to be authored by 'Claude Flow Team' in the YAML frontmatter, which contradicts the actual author 'Cornjebus.' This misleading metadata aims to gain unearned trust from the user. \n
Indirect Prompt Injection (Category 8): \n
Ingestion points: Untrusted data enters the agent context via the objective argument in npx ... spawn commands and the createTask API (e.g., 'Build microservices architecture'). \n
Boundary markers: No delimiters or safety instructions are used to separate user-provided objectives from the system commands. \n
Capability inventory: The skill possesses the ability to spawn workers, execute shell commands, and interact with a SQLite database. \n
Sanitization: There is no evidence that user-provided objectives are validated or sanitized before being incorporated into command strings or task descriptions.

hive-mind-advanced