The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to analyze source code and metadata from external, untrusted cryptocurrency projects (e.g., scanning src/ directories for rug-pull patterns). This creates a vulnerability surface for indirect prompt injection, where malicious instructions hidden in the comments or data of the target token could attempt to subvert the agent's audit logic.
Ingestion points: The agent reads file content via shell commands such as grep and blockchain inspection tools.
Boundary markers: The skill does not provide specific delimiters or instructions to ignore instructions embedded within the analyzed data.
Capability inventory: The skill utilizes shell command execution for searching and blockchain CLI tools for on-chain state verification.
Sanitization: No sanitization or filtering logic is provided for the data read from external sources.
[COMMAND_EXECUTION]: The skill provides a series of shell commands for the agent to execute, including grep for pattern matching and Solana CLI tools (spl-token, solana account) for inspecting token state. These commands are applied to external data, which is an expected part of the auditing workflow but involves logic execution based on untrusted inputs.

meme-coin-audit