The Agent Skills Directory

CREDENTIALS_UNSAFE (HIGH): The skill explicitly instructs the agent to persist generated wallet private keys to local files such as ~/.monad-wallet or .env. This practice facilitates the exposure of sensitive credentials to other processes, subsequent agent sessions, or potential exfiltration if the filesystem is compromised.
EXTERNAL_DOWNLOADS (MEDIUM): The skill directs the agent to fetch instructions from an external URL (https://docs.monad.xyz/llms.txt). This allows external, unverified content to influence agent behavior at runtime, effectively acting as an unverifiable dependency for instructions.
DATA_EXFILTRATION (LOW): The skill utilizes curl to interact with https://agents.devnads.com, a domain not present in the trusted or whitelisted sources list. While intended for faucet and verification services, it creates a potential channel for data transmission to a non-validated endpoint.
COMMAND_EXECUTION (LOW): The skill relies on shell execution for core tasks using forge, cast, and curl. While these are legitimate tools for blockchain development, their use on external or dynamically generated data increases the risk profile.
Indirect Prompt Injection (LOW): The skill is vulnerable to tool output poisoning due to its reliance on external documentation for decision-making.
Ingestion points: File SKILL.md (via instructions to fetch https://docs.monad.xyz/llms.txt).
Boundary markers: Absent; no instructions provided to the agent to treat external content as untrusted.
Capability inventory: Extensive subprocess calls (forge, cast, curl) and file-writing capabilities (~/.monad-wallet).
Sanitization: Absent; the agent is not instructed to validate or sanitize the content fetched from external documentation.

monad-development