ooda
Skill6 starsChanged 29 days agoMITView on GitHub
Runs John Boyd's OODA decision cycle as a working protocol on a task the user has explicitly asked to work that way. Requires an Orientation Block — the current model of the situation, what changed since the last cycle, and what evidence would falsify it — before any decision is committed, then returns to Observe after acting. Reasons through a structured-thinking tool, one pass per phase. Invoke only on explicit request — "run OODA", "OODA this", "/ooda", "use the OODA loop", "give this a method" — or when the user opens a long multi-stage piece of work and asks for a method to run it. This skill does not select itself; if it was not asked for, it does not run. For a failure that already happened use five-whys. For pre-deploy robustness review use chaos-engineer.
5 other projects use a version of this.
Adopted by
Most recently taken up by
- dvy1987/agent-loomtheir own version, first seen Nov 15, 2026
- Sumit189/claude-codestheir own version, first seen Nov 1, 2026
- ironyjk/strategy-frameworkstheir own version, first seen Oct 17, 2026
- vansvost-arch/claude-skillstheir own version, first seen Oct 15, 2026
- neurofoo/agent-skillstheir own version, first seen Jan 10, 2026
Other versions5 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 5
- neurofoo/agent-skills★ 111same name
- ironyjk/strategy-frameworks★ 14same name
- dvy1987/agent-loom★ 4same name
- Sumit189/claude-codes★ 1same name
- vansvost-arch/claude-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository30 other files it gives its agents.
- adversarial-re-readSkill
- adversarial-reviewSkill
- build-estimateSkill
- build-evaluateSkill
- build-executeSkill
- build-planSkill
- build-polishSkill
- build-prdSkill
- build-probeSkill
- build-validateSkill
- excel-connector-governanceSkill
- five-whysSkill
- insight-builderSkill
- insight-crystallizerSkill
- insight-promotionSkill
- multi-sessionSkill
- notion-governanceSkill
- prd-writerSkill
- promote-to-toolingSkill
- quality-reviewSkill
- ralph-implementSkill
- ralph-loopSkill
- search-governanceSkill
- session-outputSkill
- skill-validatorSkill
- task-planSkill
- verifySkill
- voice-profile-testSkill
- why-diagnosticSkill
- writing-qualitySkill
The file
We split this skill into the rules it gives an agent
11 rules. 0 are common, 11 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 5 edited versions we read keep these.
11
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 31
Every phase transition opens with a call to a structured-thinking tool (the sequential-thinking MCP, or whatever the host repo uses) naming …
- Distinctive: line 36
If that tool errors, retry once — a malformed call is the common cause, not a dead server.
- Distinctive: line 41
Do not halt. A decision loop that refuses to run when tooling is unstable refuses exactly when it is most useful.
- Distinctive: line 57
Write an Orientation Block with exactly these four fields:
- Distinctive: line 64
No decision is committed until that block exists.
- Distinctive: line 74
Do the smallest thing that produces new information.
- Distinctive: line 79
Orient to Observe: When CONTRADICTS is non-empty, the current model is stale.
- Distinctive: line 80
Do not push forward through Decide and Act on a model the evidence just broke.
- Distinctive: line 83
Orient to Act: When the situation matches a pattern already well understood, go straight to Act and skip a formal Decide.
- Distinctive: line 93
Stop looping when any one of these holds:
- Distinctive: line 99
The remaining decisions are reversible and cheap: Stop cycling, just do them.
“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
OODA
A decision cycle for work whose conditions change while you are doing it. Observe what is true, orient on what it means, decide, act, then observe again because acting changed the situation.
The version most people carry — four boxes in a circle, run them in order, go faster — is not the model Boyd built, and running that version produces ceremony. Boyd drew the loop once, in The Essence of Winning and Losing (1996), as a network whose arrows run several directions, including two links that leave Orient and skip Decide entirely. Orient is the center of the thing. Speed is not the point.
Gate — run this before anything else
Was this skill explicitly asked for? The user named OODA, typed /ooda, or opened a long
multi-stage piece of work and asked for a method to run it.
- No — say so in one line and stop. Do not run the cycle because a task felt uncertain. This skill is invoked, not selected.
- Yes, but the work has known steps and stable inputs — say that, name the two or three steps, and stop. A decision cycle on work that needs no decisions is the failure mode this design exists to avoid.
- Yes, and conditions will move — continue.
Structured thinking is the reasoning substrate
Every phase transition opens with a call to a structured-thinking tool (the sequential-thinking MCP, or whatever the host repo uses) naming the phase and its input. The phases are numbered thoughts in a traceable chain, not headings in a chat reply. The Orientation Block is written inside a thought, which is what makes it auditable rather than decorative.
If that tool errors, retry once — a malformed call is the common cause, not a dead server. If it is genuinely unavailable, run the cycle inline and stamp every reply and artifact:
OODA DEGRADED — inline reasoning; structured thinking unavailable: <verbatim error>Do not halt. A decision loop that refuses to run when tooling is unstable refuses exactly when it is most useful. The stamp is the requirement. Reasoning inline is acceptable; reasoning inline without a record is not.
The four phases
Observe
Gather what is true right now. Every observation carries its source. An observation with no source is an inference and gets labeled one. Separate what changed since the last cycle from what has been true all along — the second category is background, the first is signal.
Orient
The phase everyone skips, because it is the one with no obvious deliverable. So it gets a deliverable.
Write an Orientation Block with exactly these four fields:
MODEL — the current understanding of the situation, one paragraph
CHANGED — what is different since the last cycle
CONTRADICTS — which of my prior assumptions this evidence breaks, or "none"
FALSIFIES — the specific evidence that would prove this model wrongNo decision is committed until that block exists. If CONTRADICTS is non-empty, the model just broke — see the Orient to Observe link below before going further.
Decide
Name the options. Name what each costs if it is wrong. Pick one. Reversible decisions get made and moved past. Irreversible ones stop and ask the user.
Act
Do the smallest thing that produces new information. The action is chosen for what it reveals as much as for what it accomplishes. Then observe again, because the act changed the situation.
The two links that make this OODA and not a checklist
Orient to Observe. When CONTRADICTS is non-empty, the current model is stale. Go back to Observe. Do not push forward through Decide and Act on a model the evidence just broke. Refining a stale frame does not fix the mismatch; only new observation does.
Orient to Act. When the situation matches a pattern already well understood, go straight to Act and skip a formal Decide. This is Boyd's implicit-guidance link, and it is what keeps easy cycles from becoming ritual. State that the shortcut is being taken and name the pattern being matched — a shortcut on a superficial match is how this link fails, and naming the pattern exposes that.
Any phase can be re-entered from any other when conditions move mid-cycle. If the cycle only ever runs 1-2-3-4, it has become the cartoon version.
Exit conditions
Stop looping when any one of these holds:
- The Orientation Block's structured fields stop moving. Compare the four fields against the prior cycle's — not the prose, which will be reworded every time and would never match. If CHANGED and CONTRADICTS are both empty two cycles running, the situation has stopped moving.
- The user's stated goal is met.
- The remaining decisions are reversible and cheap. Stop cycling, just do them.
- Three cycles with no new information. Report being stuck — what was tried, what is still unknown, what would break the tie. Do not run a fourth.
State which condition fired.
Output
A short report, not a transcript — the thinking chain is already the trace. Give the cycles run, the final Orientation Block, the decisions made with their reasons, what was acted on, and which exit condition fired.
Known limitation
The gate at the top is prose, and prose gates are honor-system. The mechanical alternative is
disable-model-invocation: true in this frontmatter, which makes the skill invocable only by explicit
request and stops it competing for selection at all. Prefer it in any repo where this skill starts
firing on work nobody asked it to structure.
Related: a negative clause in a description ADDS the tokens it means to exclude, because skill selection is similarity matching rather than logic. That is why the boundary above is stated positively and by naming sibling skills, not by listing topics to avoid.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

