evernote-performance-tuning
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection as it processes and caches untrusted data from the Evernote API and user-provided search queries.\n
- Ingestion points: Search queries in
CachedEvernoteClient.searchand note metadata retrieved vianoteStore.findNotesMetadatainreferences/implementation-guide.mdingest external data into the agent's context.\n - Boundary markers: No explicit instruction delimiters or markers are used to separate untrusted data from agent instructions when processing API results.\n
- Capability inventory: The skill utilizes file system tools (Read, Write, Edit, Grep) and network libraries (
evernote,ioredis), which could be leveraged if an injection is successful.\n - Sanitization: No sanitization is performed on the content of search results or note metadata beyond hashing queries for cache keys.\n- [DYNAMIC_EXECUTION]: The skill uses JavaScript Proxies to dynamically wrap and instrument API calls for performance monitoring.\n
- Evidence: The
instrumentmethod in thePerformanceMonitorclass withinreferences/implementation-guide.mdcreates aProxyaround thenoteStoreobject to track call durations and errors at runtime.
Audit Metadata