hexagonal-architecture
Skill1 starsChanged 3 months agoMITView on GitHub
Design and implement software using Hexagonal Architecture, aka the Ports and Adapters pattern (Alistair Cockburn): isolate business logic in a technology-free core that talks to the world only through purpose-named ports with swappable adapters, so the same application runs unchanged under tests, UIs, and production infrastructure. Use this skill whenever the user mentions hexagonal architecture, ports and adapters, decoupling business logic from a database/framework/broker, swappable or pluggable backends, in-memory fakes, testing the core without infrastructure, SPI or extension-point design, or asks to design, implement, review, or refactor an application or SDK toward any of these — even if they never name the pattern.
27 other projects use a version of this.
Adopted by
Most recently taken up by
- ralvarezdev/ralvaskillstheir own version, first seen Nov 16, 2026
- AndresArreguy/ecc-pi-porttheir own version, first seen Nov 16, 2026
- ralvarez-ai/ralvaskillstheir own version, first seen Nov 16, 2026
- Lukk17/agent-standardstheir own version, first seen Nov 13, 2026
- Rojios/everything-claude-codetheir own version, first seen Nov 8, 2026
- pankaj-mahaur/Agent-Daemontheir own version, first seen Nov 6, 2026
Used with
Projects with it also use layered-architecture, domain-driven-design and 4 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
dkprogSkilllayered-architecture
Apply when structuring a backend service, creating new modules, or deciding where to place business logic. Ensures separation into transport, domain, and data access layers with correct dependency direction.
Also in 3 projects that use this
jarrodmedranoSkilldomain-driven-design
This skill should be used whenever domain modeling is taking place. It provides specialized guidance for type-driven and data-driven design based on Rich Hickey and Scott Wlaschin's principles. The skill helps contextualize current modeling within the existing domain model, identifies inconsistencies, builds ubiquitous language, and creates visualizations (Mermaid, Graphviz/DOT, ASCII diagrams) to communicate domain concepts clearly. Use this skill when designing types, modeling business domains, refactoring domain logic, or ensuring domain consistency across a codebase.
Also in 4 projects that use this
harrissohail6Skillclean-architecture
Clean architecture and SOLID principles. Auto-activates when designing systems or refactoring code structure.
Also in 3 projects that use this
GoogleSkillimprove-codebase-architecture
Scan a codebase for deepening opportunities, present them as a visual HTML report, then grill through whichever one you pick.
Also in 5 projects that use this
MicrosoftSkillgrill-with-docs
Grilling session that challenges your plan against the existing domain model, sharpens terminology, and updates documentation (CONTEXT.md, ADRs) inline as decisions crystallise. Use when user wants to stress-test a plan against their project's language and documented decisions.
Also in 4 projects that use this
GoogleSkillgrill-me
A relentless interview to sharpen a plan or design.
Also in 3 projects that use this
Other versions27 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 27
- rrezartprebreza/spring-boot-skills★ 360same name
- yamcodes/arkenv★ 151same name
- Fmarzochi/EGC★ 61same name
- fatonh/symfony-skills★ 35same name
- zernie/vigiles★ 15same name
- nledford/engineering-review-board★ 11same name
- gongyijie85/dsh-ecc★ 8same name
- ralvarez-ai/ralvaskills★ 4same name
- ralvarezdev/ralvaskills★ 4same name
- kayaman/skills★ 3same name
17 more not listed
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 repository8 other files it gives its agents.
- codexSkill
- diataxis-docs-writerSkill
- dsl-designSkill
- google-adk-agent-builderSkill
- google-adk-architectureSkill
- open-source-vendoring-best-practicesSkill
- python-sdk-designSkill
- sdk-api-designSkill
The file
We split this skill into the rules it gives an agent
5 rules. 0 are common, 4 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 26 edited versions we read keep these.
4
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 52
If the subject is a library or SDK rather than a deployable application, also read references/sdk-design.md — the pattern maps cleanly but …
- Distinctive: line 79
No technology inside the hexagon: No SQL, no HTTP types, no framework imports, no vendor SDKs.
- Distinctive: line 131
The pattern prescribes nothing inside the hexagon: No layers, no domain/application-service split, no CQRS, no DDD — those are separate, …
- Very specific: line 223
Real drivers + fakes: Add the UI/REST/CLI adapters against the still- fake back end (also the demo configuration).
- Distinctive: line 254
If not → invariant 7, and this is the finding to fix first.
“Common” is measured: kept by at least 6 in 10 of the 26 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
Hexagonal Architecture (Ports and Adapters)
This skill teaches you to produce designs and implementations that follow Alistair Cockburn's Ports and Adapters pattern as its author defines it — not the layered folklore that circulates under its name. The pattern's intent, in Cockburn's words: "Allow an application to equally be driven by users, programs, automated test or batch scripts, and to be developed and tested in isolation from its eventual run-time devices and databases."
The core move: treat the application as one self-contained component with no knowledge of any technology. Everything external — humans, tests, other programs, databases, brokers, mail servers — connects only through ports (purpose-defined interfaces the application owns) via adapters (one per technology, swappable). A test double is just another adapter, which is why a correctly built hexagon runs its full regression suite with no UI and no database.
When to apply it: applications and SDKs with real external dependencies and a lifespan — anything that must be testable without its infrastructure or survive a technology swap. When not to: small scripts and throwaway tools; the constructor parameters, interface declarations, and configurator are overhead that short-lived code never repays. Say so instead of applying the pattern ritually.
First, decide the mode
- Design mode — the user wants an architecture proposal. Follow Design workflow; deliver the design document, no code beyond port signatures.
- Implementation mode — the user wants working code. Do the design workflow first (compressed if the design is obvious), then follow Implementation workflow.
- Review mode — the user has existing code to audit or refactor. Follow Review checklist, report violations with locations, and propose the smallest refactor sequence that restores each invariant.
If the subject is a library or SDK rather than a deployable application, also read references/sdk-design.md — the pattern maps cleanly but the roles shift (the host application becomes the configurator, and test doubles become shipped product).
Vocabulary — use it precisely
Correct terminology is half the pattern; most published mistakes begin as vocabulary drift. These are Cockburn's terms:
| Term | Meaning |
| Hexagon / application | The business logic, free of all technology references. The shape is arbitrary — six sides mean nothing; the hexagon just gives room to draw many ports and breaks the top/bottom layered picture. |
| Driving (primary) actor | External party that initiates: a human at a UI, a test suite, a cron job, another program. |
| Driven (secondary) actor | External party the application calls: a database (repository — two-way) or a mail server (recipient — one-way). |
| Driving port | An interface the application implements and offers — its API, one purposeful conversation of use cases. |
| Driven port | An interface the application declares and calls — what it requires from the world (its SPI). |
| Driver adapter | Translates a driving technology into port calls: REST controller, CLI, GUI, test fixture, queue consumer. |
| Driven adapter | Implements a driven port with a technology: SQL repository, SMTP notifier, in-memory fake, recording spy. |
| Configurator | The one place that knows concrete adapters and wires everything at startup (aka composition root). |
The invariants
These nine rules are what make an implementation be the pattern. Each exists for a reason; when you must trade one off, say so explicitly rather than silently drifting.
- No technology inside the hexagon. No SQL, no HTTP types, no framework imports, no vendor SDKs. One leaked import couples the core to a technology and quietly kills the ability to test and swap — the two benefits the whole pattern exists to buy.
- The application owns every port definition — on both sides. Driven ports especially: the interface belongs to the caller, not the implementation. The persistence port looks the way the application's logic needs it to look, never the way some database driver happens to look. An interface shaped around one technology is a leaked adapter, and every other implementation will fight it.
- Name ports for their purpose: "for doing something".
ForParkingCars,ForObtainingRates,ForNotifyingSubscribers. Purpose naming keeps ports organized around conversations rather than nouns, and makes it obvious when a proposed port is really a technology in disguise (ForAccessingPostgresfails the test on sight). - Every port must plausibly support at least two adapters — and one of them is usually the test double. This is Cockburn's own acid test for whether something deserves to be a port. A "port" with exactly one conceivable implementation is just indirection. The inverse error is worse: one-technology-per-port inverts the pattern's entire point, which is that technologies substitute behind a stable port.
- All source dependencies point at the hexagon. Driver adapters depend on driving ports; driven adapters depend on (implement) driven ports; the hexagon depends on nothing outside itself. If the build system can enforce this (import rules, module boundaries), have it do so — leaks recur when nothing structural prevents them.
- Only the configurator knows concrete adapters. The hexagon receives driven adapters as constructor parameters (dependency injection) or asks a broker for them (dependency lookup — use when the choice varies per call, e.g. per-tenant storage). Test cases are their own configurators; production gets one startup module. Concrete adapter names appearing anywhere else are wiring leaks.
- Tests attach at the ports, on both sides. A test driver pushes the driving ports while test doubles answer the driven ports — full system behavior with zero infrastructure. Cockburn: a claimed Ports and Adapters implementation with no tests at the boundary hasn't implemented the pattern, it has drawn the diagram. Treat test doubles as first-class adapters that ship and version like the production ones.
- One hexagon per application — do not nest. The boundary sits exactly at the technology edge. Nested hexagons multiply boundary tests until the team stops maintaining them, at which point the inner boundaries stop being real. Internal structure can still use components with provided/required interfaces; just don't call them hexagons or give them the full ceremony.
- The pattern prescribes nothing inside the hexagon. No layers, no domain/application-service split, no CQRS, no DDD — those are separate, optional decisions. Don't present them as part of this pattern, and don't reject them either; the hexagon's interior is the user's to organize.
Design workflow
Work through these steps in order; each feeds the next.
- Identify the purposeful conversations. List everything the system talks to, then group by purpose, not technology. "Weather data arrives by RSS, HTTP, and telemetry" is one conversation (collecting weather data) with three adapters — not three ports. Expect a small number: two to four ports is Cockburn's stated norm; one port per use case and a single left/right pair are both named anti-patterns.
- Classify each conversation as driving (who initiates: them) or driven (who initiates: the application), and driven ones as repository or recipient. When one external system appears on both sides, that's two ports — the split follows the conversation, not the box.
- Name every port in "for doing something" form and give each 2–6 function signatures in domain terms only (domain types in, domain types out — no rows, requests, or wire formats).
- Enumerate adapters per port, minimum two, always including the test instrument: test drivers on driving ports, doubles (fake / stub / spy) on driven ports. If a port cannot name a second adapter, merge it back — see invariant 4.
- Plan the configurator: which wiring configurations exist (tests+fakes, demo, staging, production), injection vs lookup per port, where the startup module lives.
Deliver the design in this shape:
## <System> — Ports and Adapters design ### Actors <driving / driven, one line each> ### Ports | Port | Side | Operations | Adapters (planned) | ### The hexagon <what lives inside; explicitly technology-free> ### Configurator <configurations and wiring mechanism> ### Test strategy <what runs with zero infrastructure; what each real adapter's own tests cover>
Implementation workflow
The code shape
Five kinds of artifact, whatever the language (sketch in Python; in dynamic languages the port interfaces may stay implicit — the discipline is identical):
# hexagon/ports/driving/for_calculating_taxes.py — the app's API
class ForCalculatingTaxes(Protocol):
def tax_on(self, amount: Decimal) -> Decimal: ...
# hexagon/ports/driven/for_getting_tax_rates.py — what the app requires
class ForGettingTaxRates(Protocol):
def tax_rate(self, amount: Decimal) -> Decimal: ...
# hexagon/tax_calculator.py — the hexagon
class TaxCalculator: # implements ForCalculatingTaxes
def __init__(self, rates: ForGettingTaxRates) -> None:
self._rates = rates # injected; concrete type unknown here
def tax_on(self, amount: Decimal) -> Decimal:
return amount * self._rates.tax_rate(amount)
# adapters/driven/fake_rates.py — first-class deliverable
class FixedRates: # implements ForGettingTaxRates
def tax_rate(self, amount): return Decimal("0.15")
# adapters/driven/sql_rates.py — SQL exists here and only here
# adapters/driver/cli.py — argv in, app.tax_on() out
# main.py — the configurator:
# app = TaxCalculator(SqlRates(connect(dsn))); cli.run(app)Build order — tests and mocks first
This sequence is the pattern's native workflow; it front-loads the payoff (finished, regression-tested logic before any infrastructure exists):
- Test drivers + fake driven adapters. Implement the whole hexagon driven by tests, every driven port answered by a double.
- Real drivers + fakes. Add the UI/REST/CLI adapters against the still- fake back end (also the demo configuration).
- Test drivers + real driven adapters. Implement and verify the SQL, SMTP, etc. adapters by driving the app from tests.
- Real + real. End-to-end; every other combination is now a deployment choice, not a code change.
Folder layout
Port declarations live inside the application package; each adapter is its
own module outside it, named for the port it serves plus its technology
(adapter-forgettingrates-sql, adapter-forparkingcars-webui,
adapter-forpaying-spy). This makes the dependency rule (invariant 5)
enforceable mechanically and keeps each technology's dependencies isolated to
its adapter.
Review checklist
Audit in this order — the cheapest checks catch the most common rot:
- Grep the hexagon for technology: SQL strings, HTTP/framework imports, vendor SDK names, serialization annotations inside core logic → invariant 1 violations, each with file:line.
- Locate every port definition. Any interface defined in an adapter's package, or shaped like a specific driver's API → invariant 2.
- Check constructor signatures in the core: concrete adapter types instead of port types → invariants 5/6.
- Find the configurator. Wiring scattered across modules, or adapters instantiating each other → invariant 6.
- Run the no-infrastructure test: can the suite exercise the driving ports with doubles on the driven ports, with nothing external running? If not → invariant 7, and this is the finding to fix first.
- Count adapters per port. Ports with one conceivable implementation (merge them) and technologies bypassing ports entirely (wrap them).
For refactors, sequence fixes to restore invariant 7 earliest — once the no-infrastructure test configuration exists, every later step is protected by it.
Going deeper
- references/pattern-deep-dive.md — read when correctness of interpretation matters: the pattern's history and names, the driving/driven asymmetry, the configurator variants, common published misconceptions and their rebuttals, costs and benefits, relation to Clean/Onion/DDD, and the Component + Strategy generalization. All claims cited to primary sources (Cockburn; Garrido de Paz).
- references/sdk-design.md — read whenever the deliverable is a library, SDK, or framework rather than an application.
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.

