The Agent Skills Directory

[COMMAND_EXECUTION]: The skill workflow involves executing a shell command via cast call that interpolates multiple variables (sender, tx.to, tx.data, tx.value, tx.gas) derived from user-provided JSON input.
[COMMAND_EXECUTION]: There are no instructions or guardrails provided to ensure that the agent sanitizes these values before execution. An attacker could provide a malicious JSON object with fields containing shell metacharacters (e.g., ;, &, |, or backticks) to trigger arbitrary command execution on the host machine.
[DATA_EXFILTRATION]: If command injection is successful, an attacker could leverage the execution context to read and exfiltrate sensitive local files, such as .env files, SSH keys, or cloud provider credentials.
[COMMAND_EXECUTION]: Ingestion points: The skill processes data from the swap-build JSON output. Capability inventory: Uses cast call to interact with the shell. Boundary markers: No delimiters or ignore-instructions warnings are present. Sanitization: No validation or escaping of external content is specified before interpolation into the command line.

swap-simulate