← ClaudeAtlas

agent-harness-backend-adoption-evallisted

Evaluate a proposed library or backend as a replacement for an existing one. Produces a structured assessment: architecture comparison, benchmark credibility, unproven risks, creative angles, and a staged recommendation. Use when a new library surfaces as a candidate to replace an embedded dependency (vector DB, cache, queue, etc.).
williamblair333/Uncle-J-s-Refinery · ★ 3 · AI & Automation · score 73
Install: claude install-skill williamblair333/Uncle-J-s-Refinery
## When to use Invoke when: - A new library is proposed as a replacement for something we currently run - The change would affect a load-bearing component (persistence, search, messaging) - You need a go/no-go recommendation with honest risk accounting ## Required inputs Before starting, gather: 1. The **existing backend's pain history** — what broke, what workarounds exist 2. The **proposed backend's source** — GitHub repo, PR, or issue link 3. Our **scale/usage profile** — row counts, concurrency pattern, access frequency ## Workflow ### 1. Prior-art check (always first) mempalace_search(query="<proposed library name> <existing backend name>", limit=5) If hits exist, surface the prior decision verbatim before continuing. ### 2. Parallel gather Run these simultaneously: - Fetch the GitHub discussion / PR: `gh pr view <URL>` or `gh issue view <URL>` - Read any migration docs linked from the PR or repo README - Check if there's a contrib backlog in MemPalace for the affected component ### 3. Read the actual implementation Don't trust only benchmarks. Read: - Core data path (how writes land, where the source of truth lives) - Concurrency story (locks, file handles, multi-process access) - Error recovery path (what happens on crash — is state recoverable?) ### 4. Produce the structured assessment Output EXACTLY these six sections (no others): **What this is** One paragraph. Name the library, its architectural pattern, and the key inversion vs. what we run today. **Why