The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a surface for indirect prompt injection by demonstrating how to process untrusted data from the frontend in backend Rust commands.
Ingestion points: Multiple functions in SKILL.md, such as read_file(path: String), open_file(path: String), and fetch_data(url: String), ingest string parameters directly from the JavaScript invoke caller.
Boundary markers: There are no boundary markers or instructions advising the agent or developer to treat the frontend input as untrusted or to ignore embedded instructions within the input.
Capability inventory: The code snippets include file system access via std::fs::read, std::fs::read_to_string, and tokio::fs::File::open, as well as network access via reqwest::get.
Sanitization: The provided code lacks any input validation, path normalization, or URL allow-listing to prevent exploitation of the provided capabilities.
[DATA_EXFILTRATION]: The skill provides code templates that facilitate data exposure. The read_file and open_file functions in SKILL.md allow the backend to read and return the contents of any file path provided by the frontend. Without path validation or restriction to a specific sandbox directory, these patterns could be used to expose sensitive system files if the frontend becomes compromised or is manipulated via indirect injection.

calling-rust-from-tauri-frontend