The AI Cynefin Decider skill for Claude

That distinction sounds academic until you see what goes wrong in practice without it. The file states the problem itself in one sentence, and that sentence is the core of this whole page: the biggest mistake in decision making is applying a Clear approach to a Complex problem. A project plan with milestones, a business case with a payback period and a steering group that checks progress every month: those are excellent instruments for an issue where the right answer exists and only has to be found. On an issue where the answer only takes shape while you are working on it, they are expensive instruments measuring exactly the wrong thing.

Cynefin distinguishes five domains. In Clear, also called Obvious or Simple, cause and effect are clear and known, and best practices exist. Processing an invoice, carrying out standard maintenance. In Complicated, cause and effect do exist, but you need analysis or expertise to see them, and several good answers are possible: good practices rather than best practices. Repairing an engine, drawing up a legal contract.

In Complex, cause and effect are only visible afterwards. No best practices exist there, only emergent practice: what works only becomes clear once you have tried it. Culture change, a product launch, education.

In Chaotic there is no visible link between cause and effect, and there is a crisis: a fire, a hack, the sudden departure of someone everything depended on. And the fifth domain, Confusion, is the honest admission that you do not know which domain you are in. The file also calls this Disorder or Aporetic, and the instruction for it is simple: split it into sub-problems first, and place those sub-problems in domains separately.

What sets the framework apart from analysis models such as SWOT or PESTEL is that it does not look at the content of your issue but at the kind of issue it is. SWOT fills four boxes with what you know. Cynefin first works out whether what you know is even enough to plan with, or whether you are better off running an experiment. That makes it a layer above the other models rather than a competitor to them.

A Complicated approach to a Complex problem. This is the most common version. You set up a programme with workstreams, milestones and SMART goals for an issue whose answer only takes shape through interaction. The result is not that nothing happens, the result is that progress gets neatly reported for twelve months on indicators that say nothing. Against that the skill sets three small, reversible experiments running in parallel, with the explicit instruction to measure change in language and stories rather than output.

A Complex approach to a Clear problem. The reverse mistake happens less often but is just as costly. You start experimenting with something a working procedure has existed for all along. The skill refuses that explicitly, because probe sense respond in Clear is on its list of things it does not do, with the reason attached that it is wasteful.

Analysis during a crisis. In Chaotic the instruction is act sense respond, in that order. Act first, understand afterwards. In the second example in the SKILL.md, a data breach where the press is already calling, step one is bringing in a crisis team within thirty minutes and step three is technically sealing the leak. The deep analysis only comes once the leak is closed, and then explicitly in Complicated mode. The pitfall the file names alongside it: do not try to understand everything first, and do not ignore the press, because that only grows the chaos.

A domain classification without reasoning. The last refusal in the file is subtle but important: never give a domain classification without explaining why. A label with no reasoning is just an opinion wearing an English word in front of it. Because the skill always includes two sentences of diagnosis, you can disagree with the classification and argue your case. That is the difference between a tool and an oracle.

What that looks like in language, shown as a diff. On the left the kind of wording most plans start with, on the right what the skill turns it into once the domain turns out to be Complex.

The file opens with a frontmatter that states when the skill should trigger. Not only on the word cynefin or the name Snowden, but also on ordinary sentences: which approach fits here, how should I approach this, is this complex or just complicated, what type of problem is this. The frontmatter also explicitly names chaotic crisis situations, transformation projects and innovation questions, plus the situation where someone gets stuck by applying the wrong approach to the wrong type of problem. Say something like that to Claude while the skill is loaded, and it triggers automatically.

After that comes the role description: senior strategist trained in Dave Snowden's cynefin framework, with Cognitive Edge, IBM Research and the Cynefin Centre named as its origin. Then follows the theoretical basis with the five domains, each with the status of cause and effect, the matching decision method and an example.

The working method has five steps, each with its own heading in the file. Step 1 is clarifying the problem: with vague input the skill asks at most three clarifying questions, otherwise it moves straight on. Step 2 is determining the domain using five diagnostic questions, with the note that a problem can have parts in different domains and that you then have to split it up.

Step 3 is the domain specific approach, and that is the table you see in FIG.02 above: a decision method, a decider, tools and the typical risk for each domain. Step 4 is three to five concrete next steps that fit the domain. Step 5 is the boundary warning: where this can tip over, and the two movements the file names specifically are an ignored Complex problem falling towards Chaotic, and a working Complicated solution that migrates towards Clear over time.

The output is fixed in five blocks: classification with domain and two sentences of diagnosis, fitting approach with decision method, decider and tools, concrete steps, domain pitfalls, and boundary warning. Below that are two fully worked examples, culture change and a data breach, both with the complete output attached. Then follows the list of five things the skill does not do, and finally the source list with five titles.

What is not in it is just as relevant. There is no prioritisation method, no calculation model, no project plan template and no implementation advice. You get a classification and a working method, nothing more. What you do after that is your own work, and if you are looking for structure there, the MECE Problem Structure Coach is the logical next step, because it does the splitting step 2 asks for neatly and without overlap.

Snowden and Boone, 2007, Harvard Business Review. A Leader's Framework for Decision Making is the article that made cynefin known outside the knowledge management world. It contains the idea that leaders have to adapt their style to the domain rather than the other way round, and it also contains the description of the boundary between Clear and Chaotic as a cliff, which section 03 above is about.

Snowden, 2002, Journal of Knowledge Management. Complex Acts of Knowing is about paradox and self-awareness in description. This is the theoretical grounding under the claim that you cannot know a complex system by breaking it into parts, because its behaviour arises from the interaction between the parts.

Kurtz and Snowden, 2003, IBM Systems Journal. The new dynamics of strategy is the article that sharpens the domains and makes the idea of sense making versus categorisation explicit. If you read only one academic source, read this one.

Cognitive Edge and The Cynefin Co. The two organisations Snowden set up, named in the file as the place where the material is maintained. The framework is not finished, after all: the naming of the first domain has itself changed, from Simple to Obvious to Clear, and you come across all three names in the literature.

If you want to broaden this thinking, three logical neighbours are close by. The First Principles Thinker tackles the assumptions under your issue before you classify it. The Pre-Mortem Analyst NL lets you look ahead to the plan you just chose failing, which is especially useful in the Complicated domain, where the right answer exists but is not found on its own. And the Red Team Analyst sets targeted pushback against it, which fits exactly the risk the file attaches to Complicated: expert lock in.

After that, using it is simple: describe your problem in plain language and ask how you should approach it. The more concrete your situation, the sharper the classification. New to this kind of file? Read what Claude skills are NL first, and then how to install Claude skills NL. Want to write one yourself, for instance for the decisions that keep coming back in your line of work? Writing a SKILL.md NL explains how to go about it.

There are also situations where you are better off leaving it alone. The first is when you are looking for a worked out solution: you will not get one. Cynefin delivers a problem type and a working method, and that is deliberate. Anyone expecting a plan ends up disappointed and never uses the framework again, when it simply does something different from what they expected.

The second is genuine routine. If you are sure a working procedure exists, the classification has already been made and the skill mostly delivers a confirmation. That costs time without adding anything. The third is a live crisis where action is genuinely needed right now. In that case it correctly points you to act sense respond, but it does not replace a crisis playbook or a person with a mandate. The file itself is clear about that: in Chaotic, someone with authority decides, now.

Another honest limit worth naming. The classification you get is a language model's judgement of your description, not a measurement. Describe your issue in the language of a project plan, and there is a good chance it gets read as Complicated. Describe the same issue in the language of people who cannot agree, and Complex comes out.

That is not a bug in this skill but a property of the framework: cynefin works on the basis of how you describe something. That is why the file demands those two sentences of diagnosis, so you can see what the classification rests on and can push back on it. Anyone who wants to be sharper on that point can have the classification made twice from two different descriptions, or have it challenged by the Inversion Thinker.

If this kind of thinking works for you, take a look at what else is ready in the library of 100 free skills NL we make available with no account needed. And if you want to talk further about what else AI can do for your business, from single skills to full automation, we simply do that in a conversation. Want to know first where you stand online: the free SEO scan shows that in a few seconds. In the guide eight decision models compared NL you can see how this model relates to the others.