The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted code and manifest files from repositories, which creates a potential surface for indirect instructions to influence agent behavior.
Ingestion points: Identifies and reads .rs files and Cargo.toml manifests in SKILL.md (Steps 1, 4, and 8).
Boundary markers: Absent. There are no specific delimiters or instructions defined to separate untrusted file content from the agent's logic.
Capability inventory: Executes shell commands such as git, grep, cargo clippy, and cargo check as part of the analysis workflow in SKILL.md.
Sanitization: Absent. File contents and metadata are processed without escaping or validation against injection patterns.
[COMMAND_EXECUTION]: The skill executes cargo clippy and cargo check on the target workspace in SKILL.md (Step 3). These standard Rust tools execute build scripts (build.rs) and procedural macros defined within the project's dependency graph, which can result in arbitrary code execution if the repository is malicious.
[EXTERNAL_DOWNLOADS]: The documentation in references/rust-project-setup/references/ci-setup.md provides examples for setting up GitHub Actions, referencing well-known and community-standard services such as dtolnay/rust-toolchain and Swatinem/rust-cache.

do-review-rust