rust-review

What it does

Performs comprehensive Rust security review for safe/unsafe boundary issues, memory safety in unsafe blocks, concurrency hazards, panic-induced DoS, FFI safety, and async runtime mistakes. Use when auditing Rust crates, services, or libraries — particularly those with unsafe, FFI, or concurrent code.

How it works

• Runs in the main conversation and coordinates subagents (rust-review-worker, rust-review-dedup-judge, rust-review-fp-judge).
• Uses a Task ledger for retries; workers and judges use tool sets declared in their frontmatter. Findings are exchanged via markdown-with-YAML files in a shared output directory.
• Phase 0 collects parameters (threat_model, worker_model, severity_filter, scope_subpath). If needed, asks the user a single AskUserQuestion for unresolved parameters.
• Phase 1 runs prerequisites to determine flags (has_unsafe, has_ffi, has_concurrency, has_async, has_packed_repr, has_fs_io) by probing the finding scope with Bash-based searches and regexes. It aborts if no Rust files exist in the finding scope.
• Phase 2 creates an absolute output_dir (defaulting to a timestamped path) and ensures findings/ and coverage/ subdirectories exist.
• Phase 3 skim-read context (READ-access to README and manifest files) and write context.md with YAML frontmatter, preparing the audit context for workers.
• Phase 4 deploys the worker Prompts and Phase 5 collects findings; Phase 6 de-dup and Phase 8 report generation occur via the dedup-judge and fp-judge subagents, with SARIF output.
• The skill enforces that only allowed tools are used and that path/command references are treated as explicit instructions for the agent workflow.

When to use it

Use for Rust applications or libraries requiring security auditing around unsafe boundaries, memory safety in unsafe blocks, concurrency hazards, panic-induced DoS, FFI safety, and async-runtime mistakes.

What it can touch

The skill coordinates with agents and Bash-based probes to discover Rust code patterns and manifests. It relies on the following tools being available to subagents: Agent, AskUserQuestion, SendMessage, TaskCreate, TaskUpdate, TaskList, Read, Write, Bash.

Caveats

• Aborts if no Rust files are found in the finding scope during prerequisites.
• The workflow expects external scripts and paths (e.g., ${RUST_REVIEW_PLUGIN_ROOT}/scripts/...) to render plans and findings; misconfiguration can lead to failures.
• The process is described as multi-phase with multiple subagents; results depend on successful coordination and correct tool availability.