The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform reconnaissance on the target codebase, searching for Rust files and dependencies using standard utilities like grep, find, and rg. It also executes internal Python orchestration scripts (e.g., build_run_plan.py, validate_artifacts.py) which are part of the skill package.
[PROMPT_INJECTION]: The skill is designed to process untrusted data from the repository being reviewed, creating a surface for indirect prompt injection.
Ingestion points: The skill reads *.rs source files, README documentation, and Cargo.toml manifests from the analyzed codebase.
Boundary markers: Audit findings are exchanged via structured Markdown files containing YAML frontmatter to delimit data from instructions.
Capability inventory: Both the orchestrator and worker subagents possess Bash, Read, and Write capabilities to perform code analysis and record findings.
Sanitization: Subagent outputs are passed through a validation script (validate_artifacts.py) and a multi-stage "judge" pipeline (deduplication and false-positive filtering) before being included in the final report.
[SAFE]: The skill demonstrates robust design practices, including explicit task bookkeeping, deterministic planning via scripts, and artifact validation. No evidence of unauthorized network exfiltration, credential theft, or persistence mechanisms was found. External references are limited to trusted or well-known technical documentation.

rust-review