/evaluate — teardown a product from a URL and decide what's worth keeping

Given one URL, produce an evidence-backed verdict plus a report. Subagents crawl the code in chunks; the calling session keeps only conclusions.

Two questions, every time

Judge both — don't collapse to "lift code":

1. ADOPT AS-IS — is the product itself worth keeping/running, slotted into your Claude Code workflow (a CLI on PATH, an MCP server, a /skill or hook pattern, a dotfile)? Judged from docs + how it runs, not a code port. Note install/runtime cost and whether it's handy at the workstation even if Worker-irrelevant.
2. LIFT — features/code/patterns worth pulling into this repo, or worth keeping as a reference snippet. The code-crawl axis.

A product can win on one, both, or neither — the verdict names which.

Operating principles