iot-anomalies
Warn
Audited by Gen Agent Trust Hub on Jul 17, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads the
@claude-flow/plugin-iot-cognitumpackage from the NPM registry at runtime using thenpxutility. - [REMOTE_CODE_EXECUTION]: Executes remote code via
npx -y -p. The use of the@latestversion identifier means the skill will always run the most recent version of the package, creating a potential vector for supply chain attacks if the package or its registry entry is compromised. - [COMMAND_EXECUTION]: Spawns a bash shell to execute the
npxcommand and thecognitum-iotCLI tool. - [PROMPT_INJECTION]: Contains a surface for indirect prompt injection via the processing of IoT telemetry data.
- Ingestion points: Telemetry data fetched by the
cognitum-iotcommand-line tool. - Boundary markers: Absent. The skill does not provide delimiters or instructions to treat the data as untrusted, increasing the risk that instructions embedded in telemetry could be executed by the agent.
- Capability inventory: Shell execution (
npx), file reading (Read), and memory storage (mcp__claude-flow__memory_store). - Sanitization: Absent. The agent directly interprets anomaly types and scores from the tool's output to perform actions such as device quarantine.
Audit Metadata