Menu
Runs the agent-guided tour of Microbus using examples. Use when the user asks to take the tour.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
TRIGGER when the user asks to upgrade the project to a newer or the latest version of Microbus, or to update the framework. Each Microbus release ships this one self-contained skill; it applies that release's single-version migration, then chains to the next release's copy of this skill until the target version is reached.
TRIGGER when user asks to create, scaffold, or initialize a new microservice.
How to choose the hostname of a new Microbus microservice. Referenced by the add-microservice and add-sql-microservice scaffolding skills (and, through their delegation to add-microservice, by add-python-microservice and import-openapi-microservice). Consult it whenever a microservice's hostname is being chosen.
Performs an architectural review of a microservice-based system built on the Microbus framework. Covers only cross-cutting, cross-microservice concerns - service boundaries, the dependency graph, coupling, cross-service consistency, data ownership, workflow composition, edge security, and system operations. Anything judgeable inside a single microservice directory belongs to the review-microservice skill and is out of scope here. Produces a structured report with findings and recommendations.
Reviews the microservices touched by a set of changes - by default the whole current feature branch versus its merge-base with main, plus any uncommitted work. Runs the review-microservice skill on each changed microservice and the review-architecture skill scoped to those microservices and their graph neighbors, then consolidates one report. Use before merging a branch or before committing working-tree changes.
Performs a thorough review of a single Microbus microservice. Checks for completeness, framework compliance, code quality, security, test coverage, documentation, API design, and data access performance. Produces a structured report with findings and recommendations.
| name | take-tour |
| description | Runs the agent-guided tour of Microbus using examples. Use when the user asks to take the tour. |
CRITICAL: Do NOT explore or analyze other microservices unless explicitly instructed to do so. The instructions in this skill are self-contained to this microservice.
Copy this checklist and track your progress:
Take the agent-guided tour:
- [ ] Step 1: Download examples
- [ ] Step 2: Extend main app
- [ ] Step 3: Docker Compose
- [ ] Step 4: Run the example app
- [ ] Step 5: Example menu (OpenAPI portal + examples; loops until the user ends the tour)
- [ ] Step 6: Telemetry
- [ ] Step 7: Stop example app
- [ ] Step 8: Stop Docker containers
- [ ] Step 9: What's next
Download the latest examples from Github.
git clone --depth 1 https://github.com/microbus-io/fabric temp-clone
rm -rf exampleservices
cp -r temp-clone/exampleservices .
rm -rf temp-clone
The example files reference github.com/microbus-io/fabric/exampleservices/... internally. Replace all those references with the local module path.
Edit main/main.go and add the following block to the main app after the block of the core microservices.
app.Add(
// Example microservices
helloworld.NewService(),
hello.NewService(),
messaging.NewService(),
messaging.NewService(),
messaging.NewService(),
calculator.NewService(),
eventsource.NewService(),
eventsink.NewService(),
yellowpages.NewService(),
browser.NewService(),
petstore.NewService(),
login.NewService(),
creditflow.NewService(),
chatbox.NewService(),
weather.NewService(),
flightbooking.NewService(),
// LLM core and providers - required by the chatbox, weather, and flight-booking demos
llm.NewService(),
claudellm.NewService(),
chatgptllm.NewService(),
geminillm.NewService(),
)
Add the appropriate imports too, again, replace all references to github.com/microbus-io/fabric/exampleservices/... with the local module path.
import (
"github.com/microbus-io/fabric/exampleservices/browser"
"github.com/microbus-io/fabric/exampleservices/petstore"
"github.com/microbus-io/fabric/exampleservices/calculator"
"github.com/microbus-io/fabric/exampleservices/chatbox"
"github.com/microbus-io/fabric/exampleservices/creditflow"
"github.com/microbus-io/fabric/exampleservices/yellowpages"
"github.com/microbus-io/fabric/exampleservices/eventsink"
"github.com/microbus-io/fabric/exampleservices/eventsource"
"github.com/microbus-io/fabric/exampleservices/flightbooking"
"github.com/microbus-io/fabric/exampleservices/hello"
"github.com/microbus-io/fabric/exampleservices/helloworld"
"github.com/microbus-io/fabric/exampleservices/login"
"github.com/microbus-io/fabric/exampleservices/messaging"
"github.com/microbus-io/fabric/exampleservices/weather"
"github.com/microbus-io/fabric/coreservices/httpingress/middleware"
"github.com/microbus-io/fabric/coreservices/llm"
"github.com/microbus-io/fabric/coreservices/claudellm"
"github.com/microbus-io/fabric/coreservices/chatgptllm"
"github.com/microbus-io/fabric/coreservices/geminillm"
)
Add the LoginExample401Redirect middleware to the HTTP ingress proxy using its Init method.
httpingress.NewService().Init(func(svc *httpingress.Service) (err error) {
svc.Middleware().Append("LoginExample401Redirect",
middleware.OnRoute(
func(path string) bool {
return strings.HasPrefix(path, "/"+login.Hostname+"/")
},
middleware.ErrorPageRedirect(http.StatusUnauthorized, "/"+login.Hostname+"/login"),
),
)
return nil
}),
Ask the user if they'd like to install NATS and the LGTM stack using Docker Compose.
If the answer is yes, do so using the following command.
docker compose -f setup/microbus.yaml -p microbus up -d
CRITICAL: Stop the Docker containers when the user ends the tour or exits the session.
docker compose -f setup/microbus.yaml -p microbus down
Change into the main directory and run the example app in the backgroun.
cd main
go run main.go
CRITICAL: Interrupt or kill the example app when the user ends the tour or exists the session.
This is the hub of the tour. There are many things to explore and the user should drive, not click "next" a dozen times. Present the numbered menu below (keep the one-line descriptions) and let the user choose by number. If the user is unsure where to start, suggest 1 (the OpenAPI Portal) for a bird's-eye view of the whole running app before drilling into individual microservices.
When the user picks an item, go to its subsection below, present it, and offer the deeper "Explore more" dive if they want it. Then return here and present the menu again. Repeat until the user chooses "End the tour" (option 0), then proceed to Step 6.
1. OpenAPI Portal - The whole running app from the outside: every service's endpoints in one interactive surface.
2. Calculator - Typed functional endpoints (RPCs), input validation, and custom types.
3. Hello - Config properties, tickers, embedded resources, and inter-service calls.
4. Messaging - Unicast, multicast, and direct addressing, plus the distributed cache.
5. Browser - Outbound HTTP requests through the egress proxy.
6. Petstore - A third-party REST API delegated one-for-one from its OpenAPI document.
7. Yellow Pages - A SQL CRUD microservice with CRUD, bulk, and REST endpoints and a web UI.
8. Login - JWT authentication and role-based access control.
9. Credit Flow - Agentic workflows: fan-out/in, forEach, goto loops, subgraphs, reducers.
10. Chatbox - LLM tool-calling, with a simulated provider and optional real providers.
11. Embedder - A Go microservice using a real Python library for its compute.
12. Weather - An LLM agent built as a durable workflow, chaining two tools to answer a question.
13. Flight Booking - An agentic workflow with a real human-in-the-loop pause (Interrupt/Resume) and a subgraph.
0. End the tour.
The openapi.core portal aggregates the OpenAPI documents of every running service into a single interactive surface, so the user can take in the whole running app from the outside before drilling into any one microservice. A good first pick.
Explain the portal to the user and present the following links for them to experiment with in their browser:
Mention that the portal is generated automatically from each service's live subscriptions - there's no hand-maintained spec file to drift. Tasks, inbound events, and outbound events are filtered out at the source so the portal shows only what's externally callable (functions, web handlers, workflow graphs).
Explore more (offer, optional): walk the user through the structure of one service's entry (tags, paths, parameters, request/response schemas) and how the JSON output would feed an LLM tool-calling client.
Return to the Example Menu.
The calculator.example microservice demonstrates basic functional endpoints (RPCs) that perform mathematical operations. It shows how to define typed request/response functions, handle input validation errors, and work with custom types like points.
Explain the microservice to the user and present the following links for them to experiment with in their browser:
Explore more (offer, optional): prepare and show an overview of the microservice's features; the user may ask to see the full implementation code of any feature (show the full code, not just signatures).
Return to the Example Menu.
The hello.example microservice demonstrates a variety of framework capabilities including configuration properties, tickers, embedded static resources, and inter-service communication. It calls into calculator.example to show how microservices interact with each other, and multicasts a ping to discover all running microservices.
Explain the microservice to the user and present the following links for them to experiment with in their browser:
Explore more (offer, optional): prepare and show an overview of the microservice's features; the user may ask to see the full implementation code of any feature (show the full code, not just signatures).
Return to the Example Menu.
The messaging.example microservice demonstrates service-to-service communication patterns - unicast (load-balanced), multicast (all peers respond), and direct addressing - as well as the distributed cache for sharing state across instances.
Explain the microservice to the user and present the following links for them to experiment with in their browser:
Explore more (offer, optional): prepare and show an overview of the microservice's features; the user may ask to see the full implementation code of any feature (show the full code, not just signatures).
Return to the Example Menu.
The browser.example microservice demonstrates how to make outbound HTTP requests using the HTTP egress proxy. It provides a simple web UI with an address bar that fetches and displays the HTML source of any URL.
Explain the microservice to the user and present the following link for them to experiment with in their browser:
Explore more (offer, optional): prepare and show an overview of the microservice's features; the user may ask to see the full implementation code of any feature (show the full code, not just signatures).
Return to the Example Menu.
The petstore.example microservice demonstrates the recommended pattern for integrating a third-party REST API: a one-for-one delegating microservice generated from the API's OpenAPI document by the import-openapi-microservice skill. Where Browser showed raw egress, Petstore shows the typed pattern - each upstream operation becomes a Microbus endpoint that forwards through the HTTP egress proxy to the real public Swagger Petstore API. Only a curated three-endpoint subset of the upstream API was imported (AddPet, GetPetById, UploadFile).
This step requires outbound internet access. The read endpoint is public, so no credential or configuration is needed.
Explain the microservice to the user and present the following link for them to experiment with in their browser:
AddPet (POST) and UploadFile (POST, binary upload) are not browser-clickable; mention they exist and that a write API would normally be gated with requiredClaims rather than left open like these public reads.
Explore more (offer, optional): point out that service.go is almost entirely the shared makeFunctionRequest/makeWebRequest helpers plus one-line handlers, and that openapispecs.json is the durable record of the upstream API. Offer to show the full implementation code of any feature.
Return to the Example Menu.
The yellowpages.example microservice demonstrates a SQL CRUD microservice scaffolded using the sequel skills. It persists Person records with fields for first name, last name, email, and birthday. It includes a full set of CRUD, bulk, and REST endpoints, as well as a web UI for testing. If a SQL database is not configured, it falls back to an in-memory store.
Explain the microservice to the user and present the following links for them to experiment with in their browser:
Explore more (offer, optional): prepare and show an overview of the microservice's features; the user may ask to see the full implementation code of any feature (show the full code, not just signatures).
Return to the Example Menu.
The login.example microservice demonstrates authentication and role-based access control using JWT tokens. It includes a login form, protected pages with different role requirements, and cookie-based session management. Try logging in with admin@example.com, manager@example.com, or user@example.com (any password works).
Explain the microservice to the user and present the following link for them to experiment with in their browser:
Explore more (offer, optional): prepare and show an overview of the microservice's features; the user may ask to see the full implementation code of any feature (show the full code, not just signatures).
Return to the Example Menu.
The creditflow.example microservice demonstrates agentic workflows - multi-step processes orchestrated by the Foreman core service. It implements a credit approval workflow with 11 tasks spanning credit verification, employment verification, and identity verification. The workflow showcases advanced patterns: fan-out/fan-in parallelism, forEach iteration over employers, conditional transitions, goto loops for borderline credit scores, subgraphs for identity verification, reducers for merging parallel results, time budgets, retries, and sleep signals.
Explain the microservice to the user and present the following links for them to experiment with in their browser:
Explore more (offer, optional): prepare and show an overview of the microservice's features; the user may ask to see the full implementation code of any feature (show the full code, not just signatures).
Return to the Example Menu.
The chatbox.example microservice demonstrates the LLM integration capabilities of Microbus. It implements a simulated LLM provider that pattern-matches math questions and uses the Calculator microservice as a tool - all without requiring a real LLM API key. The demo showcases the full tool-calling flow: the LLM service resolves tool schemas from OpenAPI, the simulated provider requests a tool call, the LLM service executes it against the Calculator, and the result is fed back to produce the final answer.
The demo page's provider dropdown has two options: Simulated (the no-key chatbox) and Real. The Real option passes provider="any" with the fast tier alias, and llm.core auto-selects whichever real provider (Claude, ChatGPT, or Gemini) has its APIKey configured - so a single key of any brand makes it work, with no code change. This is the provider-portability path: the example asks for a capability tier, not a specific vendor model.
Offer to set up a real provider (optional). Ask the user whether they want to try a real LLM, and if so, to paste one API key from any of the three providers. Detect which provider it belongs to from the key prefix and write it to that provider's block in the git-ignored config.local.yaml:
sk-ant-... -> claude.llm.coreAIza... (or AI...) -> gemini.llm.coresk-... (not sk-ant-) -> chatgpt.llm.core# example: a Claude key was provided
claude.llm.core:
APIKey: sk-ant-...
If the prefix is ambiguous, ask the user which provider the key is for. Only one provider needs a key for the "Any configured provider" options to work.
Skipping this step is fine - the simulated option remains fully functional without any API keys. Mention that real-provider calls are billable.
Restart the example app after adding the key. Microbus reads config.local.yaml only at startup (the configurator loads it in OnStartup; it does not watch the file), so a key added while the app is running does not take effect until restart. Restart the app before choosing the Real option, or the request resolves to no configured provider and errors.
Explain the microservice to the user and present the following link for them to experiment with in their browser:
Suggest the user try questions like "What is 6 times 7?", "How much is 100 divided by 4?", or "Calculate 15 plus 28". Also try "Hello!" to see how it handles unrecognized questions. The dropdown lets them swap between the simulated chatbox and the Real option (which needs a key set above and an app restart).
Explore more (offer, optional): prepare and show an overview of the microservice's features; the user may ask to see the full implementation code of any feature (show the full code, not just signatures).
Return to the Example Menu.
The embedder.example microservice demonstrates how a Go microservice can use a real Python library (here, sentence-transformers) for its core compute. It loads the all-MiniLM-L6-v2 model in an in-process Python virtual environment via the github.com/microbus-io/pyvenv module and exposes typed Go endpoints that delegate to it via svc.venv.CallAndAwait.
The microservice is already running as part of the example app. The Python venv is not started at OnStartup; the user explicitly triggers it from the demo page so the multi-second cost of pip install sentence-transformers and the model download is obvious rather than hidden. Skip this step if the user does not have python3 on $PATH or does not want to download ~80MB on first run.
Present the demo link:
Once the venv is ready, these direct REST calls also work:
Explain the lifecycle to the user:
OnStartup constructs the *pyvenv.Venv but does not start it. The bus accepts the microservice for control-plane traffic immediately; the Python subprocess is not spawned until the user clicks Initialize.Embed and Similarity are subscribed with sub.Manual() and sub.Tag("python") so they stay off-bus until pyvenv's LivenessCallback fires StateReady. The microservice's onVenvLiveness handler activates the python-tagged subs at that point. Before that, calls to /embed or /similarity get a clean 404 ack-timeout.Demo, DemoInit, and DemoStatus are not manual and are reachable immediately. The page long-polls DemoStatus to surface tailed pip install and Python stdout/stderr while the venv warms up.LivenessCallback fires StateDied; the microservice deactivates Python subs and schedules a fresh Start in the background. Recovery is fast since the on-disk venv is reused.OnShutdown calls svc.venv.Close(ctx) to kill the subprocess and clean up the on-disk venv.Explore more (offer, optional): walk through service.go, python.go, and service.py (at the microservice's root, alongside the Go files) to show how the Go side delegates to Python via svc.venv.CallAndAwait and how the manual-subscription pattern is wired.
Return to the Example Menu.
The weather.example microservice is the suite's canonical answer to "how do I build an agent?" Its headline form is a workflow: AskAgent is a single-node graph whose one task runs llm.core's ChatLoop as a subgraph, exposing the microservice's own LatLng and Forecast endpoints as the model's tools. Ask it a question and the model geocodes the location with one tool, fetches conditions with the other, and composes a natural-language answer - the full tool-calling loop, which as a workflow the foreman drives so each step is durable and independently budgeted. The tour's clickable /ask runs that same loop synchronously (see below) so a browser gets an immediate reply.
Like the Chatbox "Real" option, this example needs a real LLM provider because the simulated provider does not do real tool-calling. If the user set up a provider key during the Chatbox step (10) and restarted the app, it already works here. If not, point them back to the Chatbox step's "set up a real provider" instructions (one key of any brand, written to config.local.yaml, then restart). Real-provider calls are billable.
Explain the microservice to the user and present the following links for them to experiment with in their browser:
ask runs the same tool-calling loop synchronously via llm.core's Chat (not the durable workflow), so the tour has one clickable URL that returns immediately. It does not touch the foreman - a workflow's own microservice never depends on the execution engine; launching AskAgent durably would be the job of whichever microservice owns the triggering event. The fact that this agent runs fine synchronously is itself the point: it is short enough not to strictly need a workflow, but the example models the workflow form for when an agent grows past a single request budget. LatLng and Forecast return deterministic mock data, so the example needs only an LLM key, not a third-party weather account.
Explore more (offer, optional): show the one-node graph in service.go (AskAgent wiring Answer to run ChatLoop as a subgraph) and the rendered ASKAGENT.mmd diagram, and contrast the workflow form with Chatbox's synchronous Chat - the same tool-calling loop, but durable, resumable, and observable step-by-step. The user may ask to see the full implementation code of any feature.
Return to the Example Menu.
The flightbooking.example microservice is the suite's showcase for the parts of the framework a synchronous LLM call cannot express: a durable graph, subgraph composition, and a real human-in-the-loop pause via flow.Interrupt / foreman.Resume. Where Credit Flow (9) is human-in-the-loop by branching and Weather (12) is a single-node agent, this workflow actually parks mid-run and waits for the traveler's decision. The BookFlight workflow searches a route, proposes one candidate flight at a time, and parks on an accept/keep-searching decision. Accepting a flight runs a separate ChooseSeatAgent child workflow as an isolated subgraph to pick a seat; keep-searching drives a goto loop back to the next candidate; an exhausted list ends with a not-booked message.
Because Interrupt/Resume is stateful and multi-round, the stop is a /demo web page rather than a clickable functional URL: a single stateless request cannot round-trip a parked flow. The page starts the flow, presents each proposed flight, and its Accept / Keep searching buttons resume the parked flow, carrying the flow key in a hidden field so nothing has to be copy-pasted.
Unlike Weather, this example is usable without a real LLM provider: only the final seat selection uses the LLM, and it degrades gracefully to the first available seat when no provider is configured, so the human-in-the-loop booking still completes end-to-end. If the user set up a provider key during the Chatbox step (10) and restarted the app, the seat is instead chosen by the model to match the natural-language preference. Real-provider calls are billable.
Explain the microservice to the user and present the following link for them to experiment with in their browser:
Suggest the user try Keep searching a couple of times to watch the goto loop propose the next candidate, then Accept to see the seat-selection subgraph and the final booking. The Execution History and Flow Diagram on the page show each task step, the interrupted park, and the nested subgraph.
Explore more (offer, optional): show the BookFlight graph in service.go (the AwaitDecision task's flow.Interrupt park and the flow.Goto keep-searching loop), the ChooseSeatAgent subgraph invoked via the typed NewSubgraph client, and the rendered BOOKFLIGHT.mmd diagram. Contrast the real Interrupt/Resume here with Credit Flow's branch-based approval. The user may ask to see the full implementation code of any feature.
Return to the Example Menu.
Reached when the user ends the tour from the Example Menu. Skip this step if the user elected not to install the LGTM stack with Docker.
Explain to the user that they can view the telemetry collected by Grafana at http://localhost:3000. Metrics and traces are visualized in dashboards and can also be viewed in the metrics drill-down app and the traces drill-down app.
Proceed to the next step.
Interrupt or kill the example app that was spun up earlier.
Skip this step if the user elected not to install NATS and LGTM with Docker in step 3.
Stop the Docker containers that were started earlier.
docker compose -f setup/microbus.yaml -p microbus down
Tell the user this concludes the tour and suggest to them that they try creating their own microservice by prompting the agent. Offer a few example prompts to get them started: