rust-security-auditor

Rust Security Auditor

Overview

This skill audits Rust backend systems as an adversarial reviewer. It is optimized for Rust web APIs, PostgreSQL-backed services, async runtimes, dependency supply chain risk, unsafe boundaries, and AI/LLM integrations. The goal is not to run a generic checklist. The goal is to predict how a capable attacker would chain small mistakes into real compromise: data theft, auth bypass, privilege escalation, denial of service, persistence, cost abuse, prompt/tool hijacking, or destructive database actions.

Think in systems, not isolated bugs. A low-severity parser panic becomes critical if it is reachable pre-auth and restarts the only worker. A harmless dynamic ORDER BY becomes SQL injection when combined with unvalidated column names. A safe Rust endpoint becomes unsafe when a blocking Diesel connection is held across .await and starves Tokio. An LLM summary feature becomes data exfiltration when retrieved documents can inject tool calls.

The final output should be a detailed markdown report with ranked findings, exploitability reasoning, affected code paths, abuse chains, remediation, and verification steps. Prefer a new timestamped audit directory named audit_DD-MM-AAAA-HH-MM/ at the backend root, with the main report at audit_DD-MM-AAAA-HH-MM/RUST_BACKEND_SECURITY_AUDIT.md, unless the user or repository asks for a different path.

When to Use

Use this skill when asked to:

• Audit Rust backend code for vulnerabilities, reliability risks, or hacker-mindset abuse paths.
• Review PostgreSQL usage in Rust: SQLx, Diesel, tokio-postgres, deadpool-postgres, bb8, SeaORM, raw SQL, migrations, row-level security, pool sizing, transactions.
• Audit Axum, Actix Web, Rocket, Warp, Tower middleware, extractors, auth, sessions, CORS, cookies, uploads, websockets, background jobs, queues, or webhooks.
• Review Rust dependency supply chain and crate security using cargo audit, cargo deny, cargo tree, lockfile checks, and malicious/typosquat crate heuristics.
• Review AI/LLM features inside Rust services: prompt injection, RAG poisoning, excessive agency, unsafe output handling, memory poisoning, tool-call abuse.
• Produce a severity-ranked markdown report, not just quick comments.