analyse-onchain
Pass
Audited by Gen Agent Trust Hub on Aug 20, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes a local Python script at
.agents/skills/crypto-onchain-data/onchain_fetch.pyto retrieve raw on-chain metrics like MVRV-Z, NUPL, and realized price. This execution is confined to the local filesystem and serves as the primary data ingestion method for the interpretation lens. - [INDIRECT_PROMPT_INJECTION]: The skill processes data originating from external cryptocurrency data providers (e.g., Glassnode, CryptoQuant) via the fetch script. While this introduces a theoretical attack surface for indirect prompt injection, the risk is minimized by the following factors:
- Ingestion points: Data enters the context via the output of
onchain_fetch.py. - Boundary markers: The instructions mandate a structured scoring system and timestamp validation ("as-of"), which act as logical boundaries, although no explicit "ignore embedded instructions" directive is present.
- Capability inventory: The skill's primary capability is executing the local Python script; it does not perform direct network operations or sensitive file writes.
- Sanitization: The skill logic performs numerical thresholding and scoring, which naturally filters non-numerical or instructional content that might be present in the data stream.
Audit Metadata