The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to install official QuickNode software and SDKs from standard registries (npm, PyPI, RubyGems, Cargo, Homebrew). These are legitimate vendor tools including @quicknode/sdk, quicknode-sdk, and the qn CLI.
[COMMAND_EXECUTION]: The skill includes extensive documentation for the qn CLI, teaching the agent how to manage endpoints, streams, and security settings. These capabilities are intended for infrastructure automation and include instructions for user confirmation and safe credential management.
[DATA_EXPOSURE]: Documentation describes the usage of QuickNode's IPFS gateway (quicknode-ipfs.com). While automated scanners flagged these as suspicious, they are legitimate vendor-controlled subdomains for content-addressed storage as defined in the vendor context.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external blockchain data via Streams and Webhooks (e.g., in references/quicknode/streams-reference.md), which constitutes an ingestion surface for untrusted data.
Ingestion points: Data enters the agent context through blockchain event streams, transactions, and webhook payloads.
Boundary markers: The skill includes explicit 'Safety Defaults' in SKILL.md, requiring the agent to use testnets by default and obtain user confirmation for any transaction or fund-spending action.
Capability inventory: The skill uses fetch for network operations and provides setup for CLI/SDK execution surfaces.
Sanitization: The instructions emphasize moving secrets to environment variables (.env) and using wallet connectors for client-side signing to ensure private keys are never exposed to the processing logic.

build-web3