The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to download executable files from a third-party source at github.com/sipeed/picoclaw using wget.
[REMOTE_CODE_EXECUTION]: The instructions guide the agent to grant execution permissions (chmod +x) and run the downloaded binary (./picoclaw-linux-arm64 onboard), which allows for the execution of arbitrary external code.
[COMMAND_EXECUTION]: The setup process relies on several shell commands, including wget, chmod, git, and make, which are used to fetch and prepare the execution environment.
[INDIRECT_PROMPT_INJECTION]: The skill allows the agent to ingest untrusted data from the web (via Brave and Tavily) and from a local workspace folder, which could be exploited to manipulate agent behavior.
Ingestion points: Web search results and files within the ~/.picoclaw/workspace directory.
Boundary markers: None identified in the instructions or configuration.
Capability inventory: The skill can perform network requests and has read/write access to a local workspace.
Sanitization: No data validation or sanitization procedures are described for external tool outputs.

picoclaw-ai-assistant