The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill manages sensitive cryptographic private keys stored on the local filesystem at ~/.a2a-wallet/. The documentation explicitly states that these keys are stored in a plain-text format and acknowledges that they are accessible to the agent runtime.
[REMOTE_CODE_EXECUTION]: The installation instructions for macOS and Linux systems recommend fetching a shell script from the vendor's GitHub repository and piping it directly into the shell (curl | sh).
[EXTERNAL_DOWNLOADS]: The skill includes an update command for downloading new CLI binaries. Additionally, the Windows installation process involves downloading an executable file from the vendor's GitHub releases page.
[COMMAND_EXECUTION]: The skill operates by executing the a2a-wallet CLI tool and its various subcommands to perform blockchain transactions and agent interactions.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes untrusted data from external 'Agent Cards' and metadata provided by remote agents during task execution.
Ingestion points: Data is ingested from external URIs via the a2a card command and through metadata in agent response streams.
Boundary markers: The instructions do not define delimiters or markers to isolate instructions embedded within external agent data.
Capability inventory: The skill has access to shell command execution, network communication, and sensitive filesystem paths containing wallet keys.
Sanitization: There is no mention of sanitizing or validating the schema and content of data retrieved from external agent endpoints.

a2a-wallet