The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (HIGH): The skill's core workflow involves running cargo check and cargo test against a user-specified codebase. In the Rust ecosystem, these operations are not side-effect-free; they execute arbitrary code defined in build.rs scripts and procedural macros. This creates a high risk of RCE if the skill is applied to untrusted repositories.
[COMMAND_EXECUTION] (HIGH): The skill utilizes several system commands including cargo, git, and jq. It performs automated modifications to the local filesystem (Cargo.toml) and executes multiple build-related commands based on the contents of the files it analyzes.
[INDIRECT_PROMPT_INJECTION] (HIGH): (Evaluation of Category 8: Ingestion of Untrusted Data) The skill is highly vulnerable to malicious instructions or behaviors embedded in the data it processes.
Ingestion points: The skill reads Cargo.toml files and project metadata from the local filesystem and external registries.
Boundary markers: None. The agent does not distinguish between project configuration and potential instructions or malicious scripts embedded in the codebase.
Capability inventory: Full subprocess execution for build tools (cargo) and file-writing capabilities on the host system.
Sanitization: There is no evidence of sanitization or sandboxing to prevent the Rust build system from executing malicious code contained within the target project's build scripts or macros.

rationalize-deps