Reviews RAG flows, context assembly, model routing, structured outputs, safety filters, provenance, fallbacks, evals, cost, and latency. Use during retrieval or answer generation changes.
80%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
Low
Low-risk findings worth noting
Fix and improve this skill with Tessl
tessl review fix ./.cursor/skills/ai-architecture-review/SKILL.mdUse this skill when reviewing or modifying the RAG (Retrieval-Augmented Generation) pipeline, document retrieval, and LLM integrations.
Follow AGENTS.md review throttling and docs/codex-review-protocol.md before starting. Do not review opportunistically or mutate files during pure review. After a completed PR review, post one Review record comment on the PR (see Records); never edit the frozen docs/branch-review-ledger.md table.
Before touching any protected retrieval/ranking surface, read AGENTS.md # RAG ranking protection and docs/rag-behaviour/ (README → behaviour-map → refuted-approaches →
safeguards). Flag the touch to the user before editing, even for incidental refactors.
Protected surfaces include src/lib/rag/**, clinical-search / retrieval-selection /
released-search-order / ranking-config, evidence and result-sort helpers, source-authority
tiering, the eval harness and golden fixtures, retrieval RPCs, and the retrieval-input
producers named in AGENTS (chunking/extractors, document-index-units, enrichment, embedding
fields, table facts, assertion tagging). The authoritative list is ragRankingPatterns in
scripts/pr-policy.mjs.
RAG impact: line when those surfaces change.rag-retrieval-reviewer / clinical-governance-reviewer agents for deep
review of protected ranking or grounded-evidence changes.Also honour AGENTS.md # API and provider confirmation boundary: never run OpenAI,
Supabase, hosted CI, or provider-backed evals without explicit confirmation.
RAG_PROVIDER_MODE).OPENAI_ANSWER_TIMEOUT_MS).BigSimmo/PsychSift
0d5a97a
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.