The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from the Solana blockchain, creating an attack surface for indirect prompt injection where malicious instructions could be embedded in token metadata or account data.
Ingestion points: Blockchain account data is fetched via connection.getAccountInfo in files such as examples/bonding-curve/buy-sell.ts and examples/swap/swap.ts to determine bonding curve states and reserves.
Boundary markers: No boundary markers are present to delimit or warn the agent about untrusted content in blockchain data.
Capability inventory: The skill has the capability to sign and send Solana transactions, read local files, and make network RPC requests.
Sanitization: The skill lacks sanitization of external blockchain data before it is processed or presented to the agent context.
[DATA_EXFILTRATION]: The skill's scripts and templates access the local filesystem to read wallet credentials from ./keypair.json. While this is standard for Solana development, it represents access to sensitive data that should be handled with care.

pumpfun