analyzing-on-chain-data

Warn

Audited by Gen Agent Trust Hub on Mar 11, 2026

Risk Level: MEDIUMCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill instructions (SKILL.md) direct the agent to use the Read tool on a configuration file containing API credentials (config/crypto-apis.env). Access to files with sensitive patterns like .env poses a risk of exposure if the agent is manipulated into revealing the content.
  • [EXTERNAL_DOWNLOADS]: Fetches metrics and market data from well-known technology services including DeFiLlama (api.llama.fi) and CoinGecko (api.coingecko.com).
  • [COMMAND_EXECUTION]: Employs a namespaced bash tool (crypto:onchain-*) for running the provided Python-based analytics CLI, providing isolation for command execution.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data from external APIs (data_fetcher.py). Boundary markers and sanitization are absent in the processing logic. Capability inventory includes bash execution (SKILL.md) and file system access (SKILL.md).
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 11, 2026, 08:48 PM