carlini-dm

carlini-dm

A Claude Code skill that turns Nicholas Carlini's research decision-making methodology into a structured framework you can invoke anytime you need to decide what to work on.

Bring one idea or ten. The framework scores each across ten dimensions, eliminates the losers, deep-dives the survivors, and delivers a ranked recommendation with concrete next actions.

Works for research, startups, engineering, product decisions, and general prioritization.

How it works

You invoke /carlini-dm and describe what you're deciding — a single idea, a list of competing options, or an active project you're not sure about.

For multiple options: Claude scores each option across all 10 dimensions in a quick elimination round, cuts the bottom half, then deep-dives the top 2-3 survivors before delivering a ranked recommendation.

For a single idea or active project: Claude walks through each dimension one at a time and delivers a verdict: Continue, Kill, Pivot, De-risk first, or Practice (craft work).

Either way, you get a direct recommendation and one concrete next action per option.

The Ten Dimensions

#	Dimension	What it filters
1	Impact	Does it matter if this succeeds?
2	Uniqueness	Can only you do this, right now?
3	Comparative Advantage	Does this play to your specific strengths?
4	Knowledge & Independence	Do you understand the landscape AND think independently?
5	De-risking	Have you tested the hard part first?
6	Collaboration Leverage	Do you have the right team?
7	Timing	Is the window open?
8	Focus	Can you state this in one sentence?
9	Communication Potential	Can this be explained compellingly?
10	Kill / Persist	Should you stop, continue, or double down?

"One excellent paper is worth a thousand mediocre ones, and takes less time to write." — Carlini

Installation

Claude Code

/plugin marketplace add moralespanitz/carlini-dm
/plugin install carlini-dm@carlini-dm

Manual (personal skill)

git clone https://github.com/moralespanitz/carlini-dm
ln -s $(pwd)/carlini-dm/.claude/skills/carlini-dm ~/.claude/skills/carlini-dm

Usage

/carlini-dm

Then describe what you're deciding. Examples:

• "I have 5 startup ideas and need to pick one"
• "Should I keep working on this project or kill it?"
• "Help me prioritize these 3 research directions"
• "Is this feature worth building?"

The Source

Adapted from Nicholas Carlini's article:

How to win a best paper award (or, an opinionated take on how to do important research that matters) — 2026-03-09

The article covers coming up with good research ideas, executing them well, writing them up clearly, and navigating what happens afterwards. This skill distills the decision-making principles from all four sections into a scoring framework that works beyond academic research.

Who is Nicholas Carlini?

Nicholas Carlini is a researcher at the intersection of machine learning and computer security, currently at Anthropic. Previously a research scientist at Google Brain (2018–2023) and DeepMind (2023–2025). Ph.D. from UC Berkeley under David Wagner.

His papers have received best paper awards at IEEE S&P, USENIX Security (twice), and ICML (three times), and have been covered by the New York Times, the BBC, Nature, Science, Wired, and Popular Science.

More at nicholas.carlini.com.

Philosophy

• Taste over throughput — one excellent decision is worth a hundred mediocre ones
• Kill fast — sunk cost is the enemy of impact
• De-risk first — test the riskiest assumption before the easiest parts
• Focus on the distribution — the outcome is a sample; control what you can control
• Ideas are cheap — share freely, execute ruthlessly, collaborate widely

License

MIT