The Agent Skills Directory

CREDENTIALS_UNSAFE (MEDIUM): The skill's primary function involves managing private keys to sign blockchain transactions. Several scripts, including examples/zk-compression/compress-spl.ts and examples/setup/example.ts, access sensitive data via process.env.PRIVATE_KEY or by reading a local ./keypair.json file. While expected for its intended use case, this pattern creates a risk of sensitive data exposure if the environment is compromised. Severity is reduced from HIGH to MEDIUM as this behavior is core to the primary skill purpose.
DATA_EXFILTRATION (LOW): The skill initiates network connections to external RPC endpoints (e.g., helius-rpc.com) to interact with the Solana network. These domains are not included in the trusted whitelist, representing a low-risk external data transfer finding.
Indirect Prompt Injection (LOW): The skill interacts with untrusted data retrieved from the blockchain which could influence agent logic.
Ingestion points: Data enters the context through RPC calls like getCompressedAccountsByOwner in examples/setup/example.ts and getCompressedTokenAccountsByOwner in examples/querying/fetch-accounts.ts.
Boundary markers: No specific boundary markers or 'ignore embedded instructions' warnings are present to isolate retrieved data.
Capability inventory: The skill has powerful capabilities across its files, including sendAndConfirmTransaction, transfer, mintTo, compress, and decompress.
Sanitization: There is no evidence of data sanitization or validation of on-chain content before it is processed or logged.

light-protocol