worker-integration
Warn
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSMETADATA_POISONINGINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructions utilize
npx agentic-flow, which triggers the download and execution of code from the npm registry. The packageagentic-flowis not a verified trusted vendor or well-known service. - [METADATA_POISONING]: The skill metadata specifies the author as
agentic-flow, which contradicts the platform-reported authorruvnet. This mismatch could be used to obscure the true origin of the skill. - [INDIRECT_PROMPT_INJECTION]: The skill processes trigger types and memory keys to coordinate agent tasks, presenting a surface for indirect prompt injection.
- Ingestion points: Trigger strings and memory key patterns in
SKILL.md. - Boundary markers: No delimiters or ignore-instructions are present for the ingested data.
- Capability inventory: Execution of shell commands via
npx. - Sanitization: No sanitization of input triggers is implemented.
Audit Metadata