The AI OODA Loop Decider skill for Claude

The intended use is a structured response to uncertainty: a price move, competing options or a decision based on an old picture of the situation. First distinguish observations from assumptions, then test your interpretation, compare at most three options and propose a permitted next step with a review point. These are instructions, not a guarantee that a model follows them or that a particular outcome results.

It is written for anyone who has to decide under uncertainty while the situation keeps moving: business owners, marketers, sales teams watching a deal tip over, teams that have to contain a crisis. The file itself also states what it is not for: no long term strategy, a strategic thinking framework belongs there, and no root cause analysis, a method such as 5 Whys may help, with evidence and broader investigation where needed. It is part of the skill library that we make freely available from the AI and automation service, with no account and no sales email afterwards.

The loop is a thinking rhythm, not a magic formula. Faster action can help in some situations, but the file does not promise that the fastest party wins or that every wrong model produces a wrong result. Inspect the evidence, consequences and authority for each proposed step.

Getting inside an opponent's OODA loop describes a tactical aim: adapt in ways that can disrupt their ability to respond. It does not guarantee that a faster decision forces a particular reaction. In FIG.01 the two-day review is an illustrative proposal; the competitor's actual cycle time is unknown.

Orient receives particular attention because Boyd's sketch shows it shaping observation, decision and action while feedback reshapes it. His 1976 essay Destruction and Creation discusses changing mental models. Check assumptions, comparisons and mismatches with reality; do not infer that Orient is always skipped, must always take longest or makes every later outcome wrong when imperfect.

The sources retained in the file include Boyd's essays and briefings, notably Patterns of Conflict and The Essence of Winning and Losing, plus work by Robert Coram, Chet Richards and Frans Osinga. Their existence does not establish why Boyd did not publish a book, nor prove improved business outcomes from this skill. No systematic review or effect size is supplied here. Treat it as a thinking aid, not empirical proof of winning.

Observe covers available facts, competitor or customer changes, recent events, measurements, unconfirmed signals and what is changing or stationary. Separate source, date and confidence: a number may also be an unverified report. Orient asks about mental models, untested assumptions, an outsider's view, hypothetical competitor choices, previous patterns, synthesis and mismatch. A competitor scenario is not a prediction, and no missing datum should be invented.

Decide asks for two or three realistic options, useful tempo, open or closed alternatives, a fallback and a small learning commitment. The original seventy-per-cent-now versus one-hundred-per-cent-next-week slogan remains explicitly unverified as a Boyd attribution, not a measured threshold or permission to accept 30 per cent error risk. Act proposes a permitted step, an observation that supports or challenges the hypothesis, involved people and a review time. A measurement is not automatic proof of success.

The tempo section asks about current rhythm, careful shortening of Orient, parallel activity, a daily check or weekly pivot, and delegated authority. These are planning options, not universal schedules. Auftragstaktik predates Boyd; the skill describes intent with constraints, authority and accountability, not a ban on instructions. The five outputs remain Observe, Orient, up to three Decide options, Act and a tempo check. Aim for three to five observation bullets without inventing facts. People, ideas, hardware is a prioritisation reminder, not a rule that tools always come last.

How an instruction file like that is put together and how to write one yourself is explained separately from this skill in writing SKILL.md NL.

The examples contrast conclusions with labelled assumptions, an open-ended discussion with a shortlist, and a vague next step with a proposed measurement and review time. These are illustrative differences, not measured claims about all brainstorms or behaviour observed in a model test. The shortened diff below follows the same fictional example context as FIG.01.

The second example concerns two app features and now follows all five required parts. Observe records the two options, indecision and missing usage, demand and cost evidence. Orient checks assumptions about users. Decide compares a small test of either feature or gathering missing evidence, with time, fallback and commitment made explicit or left unknown. Act proposes an authorised observation and review criterion. Tempo seeks useful feedback without assuming the easiest feature to disprove is the best choice.

After that, using it is simple: describe the situation as it is, mess and all. What you know, what you suspect, what has just changed. The file requests source labels, uncertainty and relevant follow-up questions; inspect the output for compliance. A step by step guide for every environment is in installing Claude skills NL.

Five limits remain explicit. First, the model includes feedback and implicit guidance, not just a sequential checklist. Second, a short-loop exercise is not sufficient strategy or risk assessment for multi-year investments. Third, safety, regulation and legal process can outweigh speed. Fourth, the cited work is not empirical proof of this skill's business outcomes. Fifth, reversibility alone does not make errors acceptable: agree exposure, consent, safety and authority, and allow evidence and specialist review for major decisions.

That last limit deserves its own sentence, because there is a second skill behind it. For big and irreversible steps you deliberately slow the loop down, the file says. But how do you know if a step is irreversible? There is a separate thinking framework for that, Jeff Bezos's Type 1 and Type 2 framework, and that sits in the Two-Way Door Decider skill NL. The two can complement each other: a reversibility check can inform caution, while the loop structures feedback. Neither replaces a full risk assessment.

For a root-cause investigation, this particular skill points to a separate method such as 5 Whys, which may itself need additional evidence and analysis. That scope boundary does not mean OODA ignores the past: previous experience and feedback are part of the model. Nor is a rapid loop a substitute for full long-term strategy.

The first phase of every loop is Observe, and that applies to your visibility too. Many businesses decide on their marketing by gut feeling, without a single hard observation about where they stand. The free SEO scan offers page-level checks that can contribute observations. Check its scope, sources and limitations; it is not a complete market diagnosis or evidence of Boyd’s endorsement. In the guide eight decision models compared NL you can see how this model relates to the others. And if you want to talk through what else AI can do for your business, from single skills to full automation, we simply do that in a conversation.