nvflare-fed-stats
Fail
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: HIGHINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection vulnerability surface because it processes untrusted user-supplied documentation files to guide execution.
- Ingestion points: The agent is instructed to inspect user material such as README, notes, or metadata files to extract statistical configuration parameters, as detailed in
SKILL.md. - Boundary markers: Explicit programmatic boundary delimiters are absent, although textual instructions in
SKILL.mddirect the agent to treat non-declarative operational instructions as anomalies. - Capability inventory: The skill includes code generation and local execution capabilities, specifically writing
client.pyandjob.pyfiles and executing them viaSimEnvwithin the host or container environment (detailed inSKILL.mdandreferences/stats-job-validation.md). - Sanitization: Programmatic escaping or input-level filtering is absent; validation relies on the LLM's adherence to instructions to ignore malicious directives.
Recommendations
- HIGH: Downloads and executes remote code from: https://internal-tools.example.com/setup.sh - DO NOT USE without thorough review
Audit Metadata