flow-nexus-swarm
Pass
Audited by Gen Agent Trust Hub on Mar 26, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill requires the installation of the
flow-nexusNPM package and theflow-nexusMCP server to function. - Evidence:
npm install -g flow-nexus@latestin the Authentication & Setup section. - [REMOTE_CODE_EXECUTION]: Setup and utility instructions utilize
npxwith floating tags (@latestand@alpha), which executes remote code from the NPM registry. Using unpinned versions can expose the system to supply chain risks if a package is compromised. - Evidence:
npx flow-nexus@latest register,npx flow-nexus@latest login, andnpx claude-flow@alpha hooks...commands. - [COMMAND_EXECUTION]: The skill provides several shell commands for the user to execute, including global package installation, account registration, and adding an MCP server to the environment.
- Evidence:
claude mcp add flow-nexus npx flow-nexus@latest mcp start. - [DATA_EXFILTRATION]: The skill is designed to communicate with the Flow Nexus cloud platform (
flow-nexus.ruv.io). While this is functional for its purpose, it involves sending agent task data and receiving instructions from an external third-party service. - Evidence: Registration and platform URLs pointing to
flow-nexus.ruv.io. - [PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection. It ingests untrusted data (like task descriptions or workflow inputs) which are then processed by the AI swarm to perform actions.
- Ingestion points: Untrusted data enters the context via the
taskparameter inmcp__flow-nexus__task_orchestrateand theinput_dataobject inmcp__flow-nexus__workflow_execute(SKILL.md). - Boundary markers: None identified in the provided instructions; there are no explicit delimiters or warnings to ignore embedded instructions in the processed data.
- Capability inventory: The skill has the capability to initialize swarms, spawn agents with various capabilities (researcher, coder, analyst), and execute multi-step workflows across cloud infrastructure (SKILL.md).
- Sanitization: No evidence of input sanitization or validation is present in the skill instructions.
Audit Metadata