The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill requires the user to provide a blockchain private key via the FITTINGS_PRIVATE_KEY environment variable. While the instructions correctly advise against hardcoding this secret and suggest using a dedicated 'spending wallet' with limited funds, managing high-entropy cryptographic keys through an agent environment is a sensitive operation.
[COMMAND_EXECUTION]: The skill uses a Node.js script (scripts/call.mjs) to perform lookups and handle payments. This script executes network requests, performs cryptographic signatures, and dynamically imports libraries (viem, @x402/fetch) to facilitate blockchain transactions.
[DATA_EXPOSURE]: The script reads its own source code from the file system (readFileSync) to generate a SHA-256 hash. This is used as an integrity check to ensure the local version of the payment logic matches the gateway's expectations before signing transactions.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted metadata from external registries (Crossref and ORCID).
Ingestion points: Data enters the context via the crossref.doi-metadata, crossref.search, and orcid.researcher-profile endpoints through the node scripts/call.mjs execution.
Boundary markers: The instructions advise the agent to compare returned metadata (title, authors, year) against the user's input, but explicit delimiters for the external data are not enforced in the script output.
Capability inventory: The agent can perform network lookups and spend USDC from the configured wallet.
Sanitization: The script parses JSON responses from the gateway but does not explicitly sanitize the text content before it is passed to the LLM context.

citation-check