understand-intent
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user input (@SRC) to perform intent classification and workflow selection. This creates a surface for indirect prompt injection where user messages could potentially influence the logic flow. However, the skill does not expose any dangerous capabilities (like network or file system access) that would make this surface exploitable.\n
- Ingestion points: User messages are scanned via
[SCAN:@SRC|typ=intent]inSKILL.md.\n - Boundary markers: None identified in the high-level DSL syntax.\n
- Capability inventory: Limited to classification and internal workflow selection; no file-system, network, or subprocess execution capabilities are defined.\n
- Sanitization: No explicit sanitization or filtering logic is visible in the pipeline definition.\n- [SAFE]: The skill uses a declarative DSL (I-Lang) for its operations. No obfuscation, hardcoded credentials, or unauthorized persistence mechanisms were identified. All identified external references (ilang.ai) belong to the skill vendor.
Audit Metadata