Skip to content

sb-dev/openquant / .agents/skills/rust-review

rust-review

Verified

Review Rust crates, services, or libraries for safe/unsafe boundary issues, memory safety, concurrency hazards, panic-induced denial of service, FFI safety, and async runtime mistakes. Use for a Rust security audit, especially unsafe, FFI, or concurrent code.

Rust security review

This is a local portable entry point adapted from the pinned Trail of Bits plugin. Attribution: Trail of Bits, https://github.com/trailofbits/skills. This adapted entry and its upstream resources retain CC-BY-SA-4.0; see the preserved licence in .claude/third-party/trailofbits/skills/LICENSE. The complete plugin source is in upstream/. Invoke this shared skill as rust-review; no named plugin agent is required. Apply repository instructions and canonical specifications first. This is a development tool, not an OpenQuant deployed research-agent bundle.

Required procedure

Read the full upstream procedure before starting. Follow all phases, evidence rules, finding schemas, retry limits, coverage gates, judging order, report safeguards, and success criteria there. Apply only the runtime substitutions below. Do not replace the full procedure with a short inline review.

  1. Resolve this skill's actual directory, including through the Claude symlink. Set RUST_REVIEW_PLUGIN_ROOT to its absolute upstream/ directory. Verify prompts/clusters/manifest.json and scripts/build_run_plan.py there. Use this pinned directory instead of upstream's environment-variable or home-directory search.
  2. Resolve the threat model, severity filter, and requested scope from the user. Ask for missing audit decisions together. Use the current model unless the user selects another available model. Record the actual model if supplied by the runtime; otherwise record unknown. Do not translate Claude model names into guessed provider equivalents.
  3. Use native file, shell, search, and delegation tools for the upstream operations. Read, Write, Edit, Bash, and Glob describe capabilities, not mandatory tool names. Use rg for the upstream regex searches. Read and review bundled scripts before running them. Keep the upstream Python 3.11+ and uv prerequisites.
  4. Maintain the upstream task ledger with host task tools, or persist equivalent task IDs, statuses, attempts, abort reasons, and worker metadata in ${output_dir}/tasks.json when those tools are unavailable. Never mark a worker complete before its artifact validation passes.
  5. Run each role from its full instruction file below, using the rendered worker prompt verbatim as task input. If generic delegation is available and permitted, supply the role file as instructions to a generic agent. Do not depend on rust-review:* agent registrations or frontmatter-enforced tool grants. Enforce the stated role boundaries yourself.
  6. If delegation is unavailable, run distinct sequential role passes. Persist each worker's artifacts before starting the next pass. Read each role file again at the role boundary. State execution_mode: sequential-role-passes in run-summary.md. State that judges were not independent agents. Do not claim independent validation. All workers and both judge passes remain mandatory.
  7. The cache primer and foreground/wave advice are Claude runtime optimizations. Use them only where the host supports their documented behavior. In sequential mode skip only the cache primer, never a review pass. With delegation, cap concurrency at the host limit and wait for every worker.
  8. Run the deduplication judge before the false-positive/severity judge. Preserve Phase 7 retry cleanup only within the current run's assigned prefixes. Run the artifact validator and SARIF/report safeguards exactly as upstream specifies. Read all final artifacts back. Report incomplete coverage, failed judges, dropped findings, and unverified severity before presenting conclusions.

Role files and resources

Interpret every relative plugin path in the upstream procedure and role files from RUST_REVIEW_PLUGIN_ROOT, not the audited source directory. The full plugin README, prompts, and script tests are preserved beneath upstream/. The original skill entry is named SKILL.upstream.md to avoid duplicate discovery. The prompt-regex test uses that filename; all other upstream contents are unchanged. The repository's third-party Trail of Bits licence applies to those files.

If the requested scope has no Rust source, stop with that fact. Do not report a Rust audit from installation checks or script tests. This port changes orchestration only; upstream runtime claims are not verification of this portable entry point.

rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit.Stars: 465kUpdated Claude CodeCodexVerifiedInstall
rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure.Stars: 465kUpdated Claude CodeCodexVerifiedInstall
rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics.Stars: 465kUpdated Claude CodeCodexVerifiedInstall
ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first.Stars: 192kUpdated Claude CodeCodexVerifiedInstall