event-modeling
npx skills add https://github.com/lexler/skill-factory --skill event-modeling
Agent 安装分布
Skill 文档
STARTER_CHARACTER = ðºï¸
What Event Modeling Produces
A set of vertical slices that fully describe a system’s behavior. Each slice is independently implementable and testable. The model uses business language throughout â no infrastructure or technical terms.
âââââââââââââââââââââââââââââââââââââââ
â Event Model â
â â
â âââââââââââââ âââââââââââââ â
â â Slice 1 â â Slice 2 â ... â
â â STATE_ â â STATE_ â â
â â CHANGE â â VIEW â â
â âââââââââââââ âââââââââââââ â
â â â² â
â â (events) â â
â ââââââââââââââââ â
âââââââââââââââââââââââââââââââââââââââ
Slice Types
Three types. Every behavior in the system fits one:
STATE_CHANGE â user does something
- Screen â Command â Event
- Command produces one or more events
- May have error events for failure paths
STATE_VIEW â system shows something
- Events â Read Model â Screen
- Read model aggregates data from one or more events
AUTOMATION â system reacts to something
- Event â Processor â Command â Event
- Background process, no user interaction
See references/slice-types.md for element rules, dependency patterns, and naming conventions.
Conversational Design Process
Work with the user through these phases. Move at the user’s pace â they might want to go deep on one slice before seeing the full picture.
Phase 1: Understand the Domain
Identify aggregates (core business entities), actors, and high-level use cases. Ask about the business processes, not technical implementation.
Phase 2: High-Level Model
Draft all slices without field details. Show the flow between them â which events feed which read models, which screens lead to which commands. This is the “map” of the system.
Format as a markdown document with one section per slice. Include slice type, aggregate, elements, and how slices connect.
Phase 3: Slice Detail
Walk through one slice at a time. For each:
- Define fields with types and example values
- Identify business rules (not simple validations â real domain rules)
- Write specifications as Given/When/Then scenarios
Phase 4: Executable Specifications
Turn specifications into approval fixture files using the bdd-with-approvals skill. That skill teaches how to:
- Design scannable fixture formats adapted to the domain
- Structure input/output for human validation
- Build parsers and formatters
Read that skill when it’s time to design fixtures. The event model specs (Given events / When command / Then events) map naturally to the approved fixture pattern.
Analyzing Existing Code
When working with an existing codebase instead of greenfield:
- Read the code to extract domain concepts
- Map existing operations to slice types (writes â STATE_CHANGE, reads â STATE_VIEW, background â AUTOMATION)
- Put code references (class names, packages) in element descriptions
- Extract specs from unit tests and comments
Output Format
Produce markdown, not JSON. Design for human readability â someone should look at the model and understand the system.
Write model artifacts to files. Ask the user where they want them (e.g., docs/event-model.md). Update the files as the model evolves through conversation.
High-Level Model
One document showing all slices and their relationships:
# [System Name] Event Model
## Aggregates
- Owner â pet owners who use the clinic
- Pet â animals registered to owners
## Slices
### Register Owner [STATE_CHANGE]
Aggregate: Owner
Screen: Owner Registration Form
Command: Register Owner â Event: Owner Registered
Error: â Owner Registration Failed
### View Owner Profile [STATE_VIEW]
Aggregate: Owner
Events: Owner Registered, Pet Registered â Read Model: Owner Profile
Screen: Owner Profile
### Notify Vet of New Patient [AUTOMATION]
Trigger: Pet Registered â Processor: New Patient Notifier
Command: Send Notification â Event: Vet Notified
Detailed Slice
Per-slice detail includes fields and specifications:
## Register Owner [STATE_CHANGE]
Aggregate: Owner
### Command: Register Owner
firstName: String â "George"
lastName: String â "Franklin"
address: String â "110 W. Liberty St."
city: String â "Madison"
telephone: String â "6085551023"
### Event: Owner Registered
ownerId: UUID â <generated>
firstName: String â "George"
lastName: String â "Franklin"
address: String â "110 W. Liberty St."
city: String â "Madison"
telephone: String â "6085551023"
### Event: Owner Registration Failed
errors: Map â {"lastName": "required"}
### Specifications
#### Successfully register with valid data
Given: (no prior state)
When: Register Owner
firstName: George, lastName: Franklin
address: 110 W. Liberty St., city: Madison
telephone: 6085551023
Then: Owner Registered
ownerId: <generated>, firstName: George, lastName: Franklin
#### Fail when required fields missing
Given: (no prior state)
When: Register Owner
firstName: George, city: Madison
Then: Owner Registration Failed
errors: {address: required, telephone: required}
#### Business rules
- All fields mandatory: firstName, lastName, address, city, telephone
- Telephone must be numeric, max 10 digits
These are defaults. Adapt the format to the domain â what matters is that a person can scan it and quickly validate correctness.
Anti-Patterns
- Technical language in element names (“insertOwnerRecord” â “Register Owner”)
- Skipping STATE_VIEW slices â every query/display is a slice
- Circular dependencies between elements
- Specs that test simple validation (“must be a number”) instead of business rules
- Jumping to fixture format before the model is understood
- Combining multiple commands in one slice â one command per STATE_CHANGE
See Also
- For executable test specifications: invoke the
bdd-with-approvalsskill - For approval testing mechanics: invoke the
approval-testsskill