The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill templates and examples repeatedly access a sensitive file path ./keypair.json to read Solana private keys for signing transactions.
Evidence found in: SKILL.md, examples/lending/borrow-repay.md, examples/lending/deposit-withdraw.md, templates/full-integration.ts, templates/lending-setup.ts, and templates/liquidity-setup.ts.
Code snippet: Keypair.fromSecretKey(Buffer.from(JSON.parse(fs.readFileSync("./keypair.json", "utf-8")))).
[EXTERNAL_DOWNLOADS]: The skill requires the installation of several external dependencies from the npm registry to interact with the Kamino protocol.
Packages: @kamino-finance/klend-sdk, @kamino-finance/kliquidity-sdk, @kamino-finance/scope-sdk, @solana/web3.js, @coral-xyz/anchor, and decimal.js.
[COMMAND_EXECUTION]: The skill provides instructions for executing shell commands via yarn cli to perform protocol operations such as deposits and account querying.
Example: yarn cli deposit --url <RPC> --owner ./keypair.json --token USDC --amount 100.
[SAFE]: The skill has an attack surface for indirect prompt injection by ingesting untrusted data from the blockchain and external oracle price feeds.
Ingestion points: KaminoMarket.load and Scope.getOraclePrices fetch external data into the execution environment.
Boundary markers: None present.
Capability inventory: The skill can execute blockchain write operations via sendAndConfirmTransaction.
Sanitization: No explicit validation of external data is performed before processing.

kamino