← ClaudeAtlas

thinking-speclisted

Use when discovery.md exists for a topic and the user wants to define how the eventual output will be judged. Trigger phrases include "let's build the rubric", "define the acceptance criteria for X", "how will we know the output on X is good", "thinking spec". Do NOT use this skill if discovery.md is missing for the topic — direct the user to thinking-discover first.
seankoji-com/claude-plugins · ★ 1 · AI & Automation · score 70
Install: claude install-skill seankoji-com/claude-plugins
# Thinking: Spec Phase Step 2 of Rensin's three-step process. The output is a grading rubric a stranger could apply, not a wishlist. ## Step 1: Resolve topic and load context List `thinking/` and confirm the slug against what is actually on disk — even if the topic was just discussed in this chat. Read `thinking/<topic-slug>/discovery.md` in full and `meta.json` for the output type. If discovery.md does not exist, stop: > "There's no brief saved for this topic yet, and the rubric has to be built against one — > otherwise we'd be defining pass/fail for work we haven't scoped. Say 'help me think through > <topic>' and we'll build the brief first." ## Step 2: Announce the mode shift with the one-shot framing This framing device is what forces criteria to become concrete. Use it near-verbatim: > "New role for this part. Imagine you're about to hand this brief to someone you've never > worked with. They get exactly one shot, and you don't get to answer follow-up questions. I'm > going to play the skeptic who has to judge their finished work cold — and press you on what > would need to be true for you to actually trust it." ## Step 3: Interview for checkable criteria Draw from the Phase 2 section of the probe bank for this output type — `templates/probes.research.md` or `templates/probes.implementation.md` in this plugin. Refuse vague answers. Every criterion must be checkable cold as pass/fail. When the user gives a vague one, show the upgrade rather than arguing: - Vag