The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the OpenSea API (such as NFT metadata and collection descriptions) that could contain malicious instructions. While the skill implements boundary markers and explicit instructions to ignore embedded directives, the presence of high-privilege write capabilities (trading, swaps) makes this a significant attack surface.
Ingestion points: Data fetched via the OpenSea REST API and WebSocket stream across all scripts in opensea-api/scripts/.
Boundary markers: Implemented in opensea-api/scripts/_response-markers.sh to wrap API responses.
Capability inventory: Includes on-chain wallet signing, marketplace order fulfillment, and token swap execution.
Sanitization: Instructions in opensea-api/SKILL.md explicitly warn the agent to treat API data as untrusted.
[EXTERNAL_DOWNLOADS]: The skill fetches and executes CLI tools and packages from the official OpenSea NPM registry (@opensea/cli, @opensea/tool-sdk). These are vendor-provided resources.
[COMMAND_EXECUTION]: The skill makes extensive use of shell scripts to perform network operations and data processing using tools like curl and jq.

opensea