CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

rag

Validate PsychSift retrieval, classification, ranking, grounding, citations, and answer behavior offline before preparing any live evaluation approval gate. Use for RAG changes, golden regressions, search quality, or answer-quality work.

Quality

88%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

RAG

  1. Run npm run workflow:rag-lab -- --write-evidence for the affected paths.
  2. Compare fixtures, manifests, tests, and prior local artifacts before changing behavior.
  3. Reproduce regressions with the smallest deterministic offline case.
  4. Run focused retrieval tests, npm run check:rag:fixtures, and npm run eval:rag:offline as selected.
  5. Enforce the RAG ranking-protection contract (ragRankingPatterns in scripts/pr-policy.mjs / ragRanking, including src/lib/rag.ts, src/lib/rag/**, src/lib/chunking.ts, src/lib/clinical-search.ts, src/lib/evidence-relevance.ts, src/lib/answer-verification.ts, supabase/migrations/*search*, and all other protected retrieval/input-producer surfaces): notify the user before editing any protected surface (even for incidental changes), include a RAG impact: PR line, and for any ranking/retrieval behaviour change require a provider-backed live before/after eval-canary pair (live canary pairs require explicit owner approval; offline eval:rag:offline / check:rag:fixtures do not substitute for the live canary).
  6. Preserve source governance, owner scope, conservative fallback, and rollback behavior.
  7. Ask before live retrieval, answer, Supabase, OpenAI, or provider-backed evaluation.
Repository

BigSimmo/PsychSift

Path

.agents/skills/rag/SKILL.md

Last updated
First committed

Is this your skill?

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.