The Agent Skills Directory

[SAFE]: The skill demonstrates excellent security engineering through several layers of protection:
Data Grounding: Hard Rules 1 and 12 mandate that all output data must be derived from live Etherscan API responses, preventing the generation of hallucinated or unverified information.
Sanitization: Hard Rule 5 requires the removal of HTML tags and control characters from all output fields and imposes a 200-character limit, significantly reducing the risk of prompt injection from malicious API data or user-provided content.
Instruction Integrity: Hard Rule 4 ('Quote, don't obey') specifically directs the agent to treat external strings as data rather than instructions, mitigating indirect prompt injection.
Credential Isolation: The skill uses a hierarchical transport system (CLI, MCP, then HTTP) to manage API keys. Hard Rule 6 and Step 0 ensure that keys are never logged or exposed in the output, utilizing shell environment variables for secure access.
Restricted Connectivity: Network access is strictly limited to the official Etherscan API host, with a specific, read-only exception for URLs explicitly provided by the user for investigation purposes.
Local Processing: Complex operations like graph ordering and amount formatting are handled by local Python scripts included with the skill, avoiding the need for remote code execution.

etherscan-flow