react-native-executorch
Warn
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- EXTERNAL_DOWNLOADS (MEDIUM): The
ResourceFetcherutility andexpo-file-systemfacilitate downloading resources from arbitrary URLs. - Evidence:
ResourceFetcher.fetch(..., 'https://example.com/llama3_2.pte')inreferences/core-utilities.mdandFileSystem.downloadAsync('https://some-audio-url.com/file.mp3', ...)inreferences/reference-audio.md. - REMOTE_CODE_EXECUTION (MEDIUM): The skill allows loading and executing
.pte(ExecuTorch) models downloaded from external sources. Maliciously crafted model files could exploit vulnerabilities in the runtime or produce unexpected tool-calling behavior. - Evidence:
useExecutorchModuleandLLMModuleload models from paths provided byResourceFetcher. - PROMPT_INJECTION (LOW): The
useLLMcomponent is vulnerable to indirect prompt injection when processing untrusted data. - Ingestion points:
llm.sendMessage(message)andllm.generate(chat)inreferences/reference-llms.md. - Boundary markers: Absent; messages are passed as raw strings or simple role-based objects without delimiters.
- Capability inventory: Includes tool calling via
executeToolCallbackand file system access throughResourceFetcher. - Sanitization: No evidence of input filtering or escaping before processing by the LLM.
- DATA_EXFILTRATION (LOW): Multiple hooks (
useOCR,useImageEmbeddings,useClassification) accept remote URLs as input. While used for legitimate inference, this behavior could be abused to perform Server-Side Request Forgery (SSRF) or confirm the existence of internal network resources.
Audit Metadata