swarm-advanced
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill guides the user to install the 'claude-flow' package using 'npm install -g claude-flow@alpha'. This is the core dependency required for the skill's orchestration patterns to function.
- [COMMAND_EXECUTION]: The skill includes several examples of shell commands for setting up the environment and executing swarm tasks, such as 'claude mcp add' and 'npx claude-flow swarm'. These are standard operational commands for the described toolset.
- [INDIRECT_PROMPT_INJECTION]: The 'Research Swarm' architecture creates a potential surface for indirect prompt injection because it is designed to ingest and process untrusted data from external sources like web searches and academic papers.
- Ingestion points: Data enters the system through tasks defined in SKILL.md such as 'web-search', 'academic-search', and 'data-collection'.
- Boundary markers: The provided scripts do not explicitly define delimiters or 'ignore embedded instructions' warnings for the external data being processed.
- Capability inventory: The skill uses various 'mcp__claude-flow__' tools which can spawn agents, execute parallel shell commands, and read/write to memory namespaces.
- Sanitization: The instruction snippets do not show explicit sanitization or filtering of the external content before it is stored in memory or analyzed by subsequent agent steps.
Audit Metadata