The Agent Skills Directory

[SAFE]: The skill implements strong defensive practices for cryptocurrency OSINT. It explicitly instructs the agent to verify checksums (Base58Check, bech32, EIP-55), compare full address strings to prevent falling for lookalike address poisoning, and treat commercial or community labels as third-party claims rather than ground truth. The 'Grading a finding' framework ensures the agent distinguishes between cryptographically confirmed facts and probabilistic inferences.
[INDIRECT_PROMPT_INJECTION]: The skill operates on untrusted external data from blockchains and third-party explorers, creating an attack surface for indirect injection (e.g., through address poisoning or deceptive contract metadata).
Ingestion points: Wallet addresses, transaction hashes, ENS names, and explorer metadata (SKILL.md, reference/address-formats.md).
Boundary markers: The skill defines a logical grading system (Confirmed vs. Probable vs. Unconfirmed) and warns that labels are 'proprietary guesses'.
Capability inventory: Involves network interaction with block explorers and pivots to other OSINT tools (reference/explorers-and-tools.md).
Sanitization: Mandates checksum verification, full-string comparisons, and manual verification of reverse resolution for naming services (SKILL.md).
[EXTERNAL_DOWNLOADS]: The skill directs the agent to utilize a variety of well-known third-party services and tools for data collection, such as Etherscan, Blockchair, and mempool.space. These interactions are standard for the skill's stated purpose of blockchain analysis.
[COMMAND_EXECUTION]: The skill mentions using local nodes (e.g., Bitcoin Core RPC) or CLI tools, but these are described as architectural recommendations for the user/agent to ensure data integrity, rather than silent or dangerous execution patterns.

follow-the-crypto