Event Model to Tasks

Translates a completed event model into implementation tasks. Each slice becomes one task with the specification as its acceptance criteria.

Input

A completed event model document (produced by event-modeling-spec skill) containing:

• Named slices, each with a pattern (Command / View / Automation / Translation)
• Specifications: Given-When-Then for commands/automations, Given-Then for views
• Swim lanes with domain assignments
• Passed completeness check

If the event model is incomplete, use event-modeling-spec first.

Task ID Convention

<TRIGRAM>-<SHORT_NAME>

• Trigram: 3 letters from the feature name (confirm with user if ambiguous)
• Short name: Short, readable abbreviation of the slice — prefer a single word when unambiguous, use two words hyphenated when needed

Slice Name	Short Name	Why
RegisterGuest	REGISTER	Unambiguous single word
InitiateBooking	BOOK	"Initiate" is ceremony — "book" is the action
RoomAvailability	AVAIL	Common abbreviation
ConfirmBooking	CONFIRM	Unambiguous
CancelBooking	CANCEL	Unambiguous
GuestProfile	PROFILE	Unambiguous
EarnLoyaltyPoints	EARN	Clear in context
OpenLoyaltyAccount	LOYALTY-OPEN	Two words needed — "open" alone is ambiguous

Infrastructure tasks use <TRIGRAM>-INFRA-<NAME> and go in Phase 0:

Task	Short Name	Why
Event persistence setup	INFRA-EVENTS	Shared foundation for all slices
Automation dispatcher setup	INFRA-DISPATCH	Required by all Automation and Translation slices

Example: HTL-INFRA-EVENTS, HTL-INFRA-DISPATCH

Rules

• One slice = one task. Do not merge. Do not split. Merging makes it impossible to correctly sequence dependent slices, eliminates parallelism within a phase, and prevents spec-impl from marking partial work as complete.
• Event names: Use exactly as named in the model. Do not rename.
• Specifications → Acceptance criteria: Copy Given-When-Then verbatim from the event model.
• No invented concepts: No aggregates, entities, or schemas not in the model.
• Infrastructure tasks go in Phase 0, marked [INFRA], kept minimal.
• Error case verification per pattern:
  • Commands: test precondition rejection (Given events missing → rejected)
  • Automations: test idempotency (run robot twice → second run does nothing)
  • Translations: test idempotency (run translator twice → second run does nothing; stub external system in tests)
  • Views: test empty/missing data response (no events → empty result or 404)

Deriving Dependencies

Trace the Given clause of each slice:

• Command depends on: any Command whose events appear in its Given clause
• View depends on: all Commands producing the events it reads
• Automation depends on: the Commands producing events for its todo-list view

Phasing

Group tasks by dependency depth:

• Phase 1: slices with no dependencies
• Phase 2: slices depending only on Phase 1
• Phase N: slices depending on Phase N-1

Tasks within the same phase can be implemented in parallel. Order tasks within a phase by: Commands first, then Views, then Automations.

Process

Always include the "How to Read These Tasks" preamble in every generated task file — never omit it. The preamble is the primary guide for developers unfamiliar with Event Modeling. It is not optional.

1. Assign trigram — Confirm with user
2. List slices — ID, pattern, swim lane
3. Build dependency graph — Trace Given clauses
4. Phase by depth
5. Write task file — Using references/output-template.md (includes developer preamble)
6. Add [INFRA] tasks in Phase 0 if needed
7. Generate Mermaid dependency graph

Output Format

The task file has two levels per task:

• Table row — for spec-impl scanning: task ID, status, one-line verification summary, blocked-by
• Detailed section — for developer/agent: Build instructions, verbatim acceptance criteria, full verification list

The Blocked by: in the Notes column is the source of truth for dependencies. The Mermaid graph is generated from it.

See references/output-template.md for the full template with preamble.

Reference

Reference	Load When
references/output-template.md	Generating the task file — template with developer preamble, table + detail format
references/tasks-example.md	Full Hotel Booking example — 18 tasks across 6 phases with Mermaid dependency graph

Common Mistakes

Mistake	Fix
Merging slices into one task	One slice = one task. Merging breaks sequencing, parallelism, and atomic completion
Renaming events from the model	Use event names verbatim — the model IS the contract
Introducing aggregates not in the model	Event Modeling doesn't prescribe aggregates
Dropping specifications from tasks	Copy Given-When-Then exactly as acceptance criteria
Applying Command error tests to Views	Commands: test rejection. Automations: test idempotency. Views: test empty response
Business logic in the automation robot	Robot just reads todo and calls command. Logic lives in handler
Preamble missing from task file	Always include "How to Read These Tasks" for the developer
Inconsistent task ordering within phases	Commands first, then Views, then Automations