The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches NFT metadata from external sources including OpenSea, Alchemy, and various IPFS gateways, which creates a surface for indirect prompt injection where malicious instructions could be embedded in trait attributes or descriptions. \n
Ingestion points: scripts/metadata_fetcher.py retrieves untrusted data from external APIs. \n
Boundary markers: No explicit delimiters or instructions are used to isolate the fetched external content in the analysis output. \n
Capability inventory: The skill uses Bash(python3:*) and Read tools. It lacks high-risk capabilities like network-write or unrestricted shell access that would facilitate significant data exfiltration. \n
Sanitization: scripts/trait_parser.py performs basic string normalization and stripping. \n- [EXTERNAL_DOWNLOADS]: The skill fetches NFT collection data and trait information from well-known technology services. \n
Evidence: Connects to api.opensea.io, eth-mainnet.g.alchemy.com, and IPFS gateways including ipfs.io, pinata.cloud, and cloudflare-ipfs.com. \n- [METADATA_POISONING]: Documentation in references/implementation.md is inconsistent with the skill's manifest and source code, referencing tools (Bash(crypto:nft-*)) and files (config/crypto-apis.env) that do not exist in the skill's environment.

analyzing-nft-rarity