The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill's code templates and examples are designed to read sensitive private key data from a local file path.
Evidence: Multiple files including templates/collateral-borrow-setup.ts, templates/lending-setup.ts, examples/lending/borrow-repay.md, and templates/full-integration.ts use fs.readFileSync("./keypair.json", "utf-8") to load Solana wallet secrets.
[COMMAND_EXECUTION]: The skill facilitate the execution of financial transactions on the Solana blockchain using provided credentials.
Evidence: The SDK methods account.deposit, account.borrow, account.withdraw, and account.repay are used throughout the documentation and templates to move assets and create debt positions.
[INDIRECT_PROMPT_INJECTION]: The skill interacts with external data from the Solana blockchain which could be manipulated to influence agent behavior.
Ingestion points: Account and bank state are fetched from the Solana network using MarginfiClient.fetch and account.reload() in all example scripts.
Boundary markers: None. The scripts do not use delimiters or instructions to ignore embedded content within blockchain metadata.
Capability inventory: The skill possesses capabilities to perform file reads (keypair.json) and execute on-chain transactions (deposit, borrow, withdraw, repay).
Sanitization: There is no evidence of sanitization or validation for data ingested from the blockchain before it is used in transaction logic.
[EXTERNAL_DOWNLOADS]: The skill fetches libraries from the Marginfi protocol and Solana ecosystem.
Evidence: SKILL.md includes instructions to install @mrgnlabs/marginfi-client-v2, @mrgnlabs/mrgn-common, @solana/web3.js, and @coral-xyz/anchor via NPM.

marginfi