The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill provides patterns for processing external file data and subprocess output while maintaining capabilities to write to the filesystem and execute arbitrary commands. This combination is a high-risk surface for indirect prompt injection.\n
Ingestion points: std::fs::read_to_string (references/file-io.md) and child.wait_with_output() (references/process-integration.md).\n
Boundary markers: Absent. No instructions are provided to help the agent distinguish between data and potential commands within the processed files or streams.\n
Capability inventory: Filesystem writes via atomic_write (references/file-io.md) and arbitrary command execution via run_subprocess (references/process-integration.md).\n
Sanitization: Absent. Technical validation (UTF-8, Serde) is covered, but there is no logic to prevent data from being interpreted as agent instructions.\n- [COMMAND_EXECUTION] (HIGH): The skill provides patterns for executing system commands using the Command API. Because the guide suggests using these for processing external file formats and process integration, there is a significant risk that untrusted data could be used to construct and execute malicious commands.\n- [EXTERNAL_DOWNLOADS] (LOW): The guide recommends installing tools like cargo-fuzz from the crates.io registry. These are standard tools but introduce a runtime dependency on external, third-party code.

rust