The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to execute shell commands such as cargo fmt, cargo clippy, and cargo test on user-provided code. The cargo test command is particularly dangerous as it compiles and runs test binaries from the codebase being audited, which can lead to arbitrary code execution if the source code is malicious.
[REMOTE_CODE_EXECUTION]: By running cargo test or cargo clippy on a project, the agent initiates the downloading and execution of external dependencies listed in the project's Cargo.toml. A malicious project could specify dependencies or build scripts that execute harmful code during the compilation or testing phase.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data (Rust source code, crates, and diffs) and use that data as input for high-privilege operations like shell execution.
Ingestion points: Rust source files, modules, crates, and pull request diffs provided by the user (SKILL.md).
Boundary markers: None identified.
Capability inventory: Shell command execution for formatting, linting, and testing.
Sanitization: None identified; the skill directly invokes tools on the provided content.
[EXTERNAL_DOWNLOADS]: The skill refers the user to a third-party GitHub repository (https://github.com/morganlinton/BeautifulRust) to obtain linting configuration files. This encourages the incorporation of external, unverified code into the development workflow.

rust-review