Free and open source|MIT licensed
Second Opinion.
A Claude skill that runs any decision through five deliberately one-sided perspectives and returns a straight verdict: GO, RESHAPE, or STOP.
It exists to break the default failure mode of AI assistants: finding five reasons to agree with whatever you were going to do anyway. The target is neither agreement nor severity. It is accuracy.
Prefer to read the code first? It lives on GitHub, and you can install it straight from there. No form required, here or there.
The email gets you a short note when the skill meaningfully changes. Nothing else, and one reply gets you off the list.
Install
Two minutes, whichever way you use Claude.
In the Claude app or claude.ai
- 1.Download the zip with the button above.
- 2.In Claude, open Settings, then Capabilities, and find Skills.
- 3.Click Upload skill and pick the zip. No unzipping needed.
- 4.Start a chat and say "I want a second opinion on this", then describe the decision.
No upload option in your plan? Open the zip, paste the contents of SKILL.md into a conversation, and ask for a second opinion. Same result.
In Claude Code
Run this once inside your project:
mkdir -p .claude/skills/second-opinion curl -o .claude/skills/second-opinion/SKILL.md \ https://raw.githubusercontent.com/cristiansimion/second-opinion/master/SKILL.md
For every project instead of one, put the same file in ~/.claude/skills/second-opinion/SKILL.md. Then type /second-opinion or ask for a second opinion in plain words.
How it works
Five voices are the easy part. Making them argue from facts is the work.
01
It asks up to three questions, writes a short neutral brief, and shows it to you so the framing can be challenged before anything runs.
02
Five personas argue from that brief in isolation, each committed to its own bias. Every factual claim needs a primary source, or the words "checked nothing".
03
A verification gate opens the citations behind every claim and cuts what does not hold. Personas invent numbers; the gate is why you can trust the ones that survive.
04
The verdict maps from the scores by rule, not by feel, and ends with the biggest risk, the honest best case, and the cheapest test you can run in 48 hours.
The council
Five perspectives, each committed to its bias.
First principles
Rebuilds the decision from its raw parts and checks whether the comparison itself makes sense.
The contrarian
Hunts for the case where the plan breaks, and for the thing the framing is hiding.
The champion
Makes the honest best case for going ahead, and says so when that case is thin.
The other side
Speaks as the person who carries the consequences, six months from now.
The realist
Goes looking for what usually happens, cites a primary source, or says plainly that none exists.
A grain of salt
Every run ends with the same warning: it is an outsider that only weighs what you tell it. Use it to think, not to decide for you.
It costs nothing, so here is the honest trade.
I built this and I give it away. If it helps you make one better call, two small things help me back: connect with me on LinkedIn, where I write about this kind of work, and if you think the skill deserves it, leave the repo a star.
Neither is required. The download works either way.