| name | rust-ml-llm-review |
| description | Rust ML/LLM review: Rig, Candle, agents, RAG, embeddings, streaming, evals, security. |
Rust ML LLM Review
Review LLM systems for layer boundaries, prompt/tool safety, provider abstraction, RAG trust, output validation, resource budgets, observability, evals, and deployment constraints.
Process
- Inspect the diff and surrounding code before applying checklist items.
- Read references/guide.md first; it is a routing index for focused reference files, then read only relevant sector files.
- Check correctness, boundaries, security, data flow, observability, tests, and deployment impact for the changed behavior.
- Report concrete findings first, ordered by severity, with file references and affected behavior.
- Call out missing tests or residual risk only when it is tied to the reviewed change.
Large PR Subagent Review
For broad or materially risky pull requests, recommend focused additional review only when independent expertise is likely to improve the verdict. Judge this from concrete risk, novelty, blast radius, reversibility, and context size—not file count or category matching alone. Small and conventional changes should remain one pragmatic pass.
When additional reviewers add value, split only by independent material surfaces. Each returns concrete evidence-based findings; the main agent removes duplicates and false positives and owns the proportional final verdict.
Coordination
Use this skill together with itsol-task-intake for ambiguous work, itsol-self-review before handoff, and focused security-* or infra-* skills when the change touches trust boundaries or deployment behavior.
