The Agent Skills Directory

[DATA_EXFILTRATION]: The skill manages authentication tokens stored in a local configuration file.
Evidence: The skill reads and writes to ~/.agentwallet/config.json to store the apiToken used for authenticated requests.
This represents standard secret management practice for its primary purpose, and the skill explicitly instructs the user to set secure file permissions using chmod 600.
[COMMAND_EXECUTION]: Relies on system tools to interact with the wallet provider's API.
Evidence: The instructions use curl for all operations, including authentication, fund transfers, and contract interactions.
[EXTERNAL_DOWNLOADS]: Retrieves skill metadata and version information from the vendor's infrastructure.
Evidence: Fetches skill.json from https://frames.ag/skill.json to check for updates.
The skill follows best practices by instructing the agent not to auto-update or overwrite files without explicit human review and approval.
[INDIRECT_PROMPT_INJECTION]: The skill includes a capability to proxy requests to external URLs, which could expose the agent to untrusted data.
Ingestion points: The /x402/fetch endpoint returns response bodies from arbitrary external target URLs into the agent's context.
Boundary markers: Present. The skill contains a specific "Data safety" section warning the agent to treat response bodies as untrusted data and not to execute instructions found within them.
Capability inventory: The agent has the ability to execute curl commands and write to its own configuration files.
Sanitization: Absent. The skill relies on the agent's adherence to the provided safety instructions to mitigate the risk of following embedded instructions.

agentwallet