The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation and internal scripts instruct the agent to download and execute official NPM packages and development tools.
Evidence: SKILL.md and scripts/balance.mjs reference @sfpro/sdk, @superfluid-finance/tokenlist, @superfluid-finance/metadata, and foundry-rs/cast.
Context: These resources are author-owned vendor resources and well-known industry tools.
[DYNAMIC_EXECUTION]: Provided utility scripts use dynamic loading to fetch protocol interface definitions from the SDK at runtime.
Evidence: scripts/abi.mjs and scripts/selectors.mjs use dynamic import() to load ABI modules from variable paths.
Context: Loading is restricted to a hardcoded map of author-vetted sub-modules within the official SDK.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface where instructions could be embedded in external blockchain or API data.
Ingestion points: scripts/balance.mjs (Super API), references/guides/api-services.md (Balance API, CMS, Accounting, Whois, Token Prices), and references/subgraphs/*.graphql (The Graph protocol endpoints).
Boundary markers: Absent. Data is retrieved and passed to the agent context in raw JSON or entity format.
Capability inventory: File system write access via scripts/selectors.mjs, network fetch capabilities in scripts/balance.mjs, and dynamic module loading in scripts/abi.mjs.
Sanitization: Absent. Results from external APIs are interpreted directly by the agent without additional validation layers.
[COMMAND_EXECUTION]: A maintenance script included in the skill is capable of writing documentation artifacts to the local directory.
Evidence: scripts/selectors.mjs uses node:fs writeFileSync to generate YAML selector files in the references/contracts/ path.
Context: This capability is used to maintain the skill's own documentation and is restricted to specific internal directories.

superfluid