The Agent Skills Directory

[PROMPT_INJECTION]: Indirect prompt injection vulnerability surface detected in the processing of external blockchain data. \n
Ingestion points: The skill fetches raw account data and transaction signatures from external Solana RPC endpoints as described in SKILL.md and implemented in decode_account.py. \n
Boundary markers: Absent; there are no instructions provided to the agent to distinguish its operational instructions from the potentially untrusted data retrieved from the blockchain. \n
Capability inventory: The skill uses curl for network requests and jq and python3 for data parsing and execution. \n
Sanitization: Absent; the blockchain data is processed and displayed without sanitization, which could allow maliciously crafted account data to influence the agent's behavior if it interprets the output as instructions. \n- [COMMAND_EXECUTION]: The skill relies on local shell commands and script execution to perform blockchain queries. \n
Evidence: Employs curl to make POST requests to RPC endpoints and jq for parsing JSON responses. \n
Evidence: Executes a local Python script, decode_account.py, to parse binary data and print it to the console.

solana