create-tasks

name	create-tasks
description	> Use when this capability is needed.

Project context: Values in angle brackets below (e.g. <scheme>, <JIRA_KEY>, <flag-key-enum>) are resolved at runtime — detect them from the project (xcodebuild -list -json for the scheme, git/gh for repo & owner, the branch name for the Jira key, a codebase search for flag/font files), or ask if they cannot be inferred. This plugin ships no per-project config.

Create Tasks from Meeting Notes / Specs

Take meeting notes, specs, or a feature description as input. Detect the platform, explore the codebase, plan the implementation, and create Jira tickets.

Usage

/create-tasks <description or paste meeting notes>

• $ARGUMENTS — Feature description, meeting notes, specs, or any context about work to be planned

---

Platform Detection

Detect the platform automatically from the current repository:

Signal	Platform
Package.swift, *.xcodeproj, Tuist/, *.swift files	iOS
package.json, tsconfig.json, src/api/, serverless.yml	Backend

If ambiguous (e.g., meeting notes mention both platforms), ask the user which platform to create tasks for — or create tasks for both if explicitly requested.

---

Jira Configuration (Shared)

• Project: <JIRA_KEY>
• Issue type: Task
• Story points field: customfield_10034
• Allowed story point values: 1, 2, 3 only
• Board: https:///jira/software/c/projects/<JIRA_KEY>/boards//backlog
• Epic: Link tasks to an epic via "parent": "<EPIC-KEY>" in additional_fields. Ask the user which epic to use if not specified.
• Assignee: Ask the user who to assign tasks to before creating tickets. Use email or display name for the assignee field.
• Sprint: Default to the current active sprint. Fetch via mcp__mcp-atlassian__jira_get_sprints_from_board (board ID <jira-board-id>, state active). Set via "customfield_10020": <sprint_id> in additional_fields. If no active sprint or the user specifies a different one, ask for input.
• Next sprint (for feature flag cleanup tickets): Fetch via mcp__mcp-atlassian__jira_get_sprints_from_board (board ID <jira-board-id>, state future) and take the first result. If no future sprint exists yet, ask the user whether to create the cleanup ticket in the backlog (omit customfield_10020) or wait.

---

Platform-Specific Configuration

iOS

• Labels: Apply all relevant labels per task:
  • Team: frontend
  • Module: e.g., explore, feed, plan-tab
  • Report: exactly one of report-feature, report-improvement-fix, report-tech-improvement, report-analytics-fix
• Story Point Scale:

  Points	Scope	Examples
  1	Trivial / mechanical change, < 1 hour	Rename a file/struct, add a field to a model, update a CodingKeys enum, delete dead code
  2	Moderate change, 1–3 hours	Update a query + response models, create a simple view, wire a new case into an existing switch
  3	Significant change, 3–6 hours	Create a new UI component from Figma, build a new repository layer end-to-end, refactor a view model with new data flow

Backend

• Labels: ["backend"]
• Story Point Scale:

  Points	Time estimate	Scope
  1	~1–4 hours	Small, focused change — add a field to a Mongoose model, update a GraphQL type, minor refactor, add a DataLoader
  2	~5–8 hours	Moderate — new service method + tests, new repository query, wiring a new GraphQL resolver with validation and permissions
  3	~9–16 hours	Significant — new domain module end-to-end, complex data pipeline, new Lambda function, major refactor across multiple layers

Hard rule (both platforms): If a task would exceed 3 points, split it. No task should be larger than 3 story points.

Trivial task rule (both platforms): If a task is purely mechanical and small enough to be part of another task (e.g., adding a feature flag, adding a constant, updating an env variable), fold it into the most relevant adjacent task. No standalone tickets for these.

Feature flag cleanup rule (both platforms): Whenever any task in the plan introduces a new feature flag (A/B test, experiment, gradual rollout, kill switch), pair it with a mandatory cleanup ticket scheduled in the next sprint. See Feature Flag Cleanup below for the full rule. Folding the flag creation into an adjacent task (trivial task rule above) does not exempt you from creating the cleanup ticket.

---

Workflow

Phase 1: Understand the Request

1. Parse $ARGUMENTS to extract:
   • What is being built or changed
   • Any GraphQL queries/mutations or API contracts provided
   • UI/UX requirements or Figma references (iOS)
   • Which modules/domains are affected
   • Any explicit non-goals or out-of-scope items
2. Extract only tasks for the detected platform. Ignore:
   • Tasks for the other platform (unless explicitly requested)
   • Process or organizational discussions
   • Non-engineering items
3. If the scope is ambiguous, note open questions and ask before proceeding.

Phase 2: Explore the Codebase

4. Launch up to 3 parallel explore agents to understand the current implementation: iOS agents:
   • Find existing files, models, views, view models, repositories related to the feature
   • Find existing patterns and components that can be reused
   • Understand the data pipeline (query → response model → mapper → repository → use case → view model → view) Backend agents:
   • Agent 1 — Domain model: Find *.model.ts and *.entities.ts in the relevant domain; check for existing migrations
   • Agent 2 — API layer: Find the relevant src/api/*-api.ts; identify existing typeDefs, resolver signatures, Joi validations, and shield permissions
   • Agent 3 — Service/repository layer: Find *.service.ts and *.repository.ts in the relevant domain; identify methods, patterns, and event bus usage
5. Read the critical files identified to verify details before planning.
6. Specifically look for:
   • Existing patterns for similar features — always follow them, never invent new conventions
   • (iOS) Reusable UI components, existing design system tokens, AppStrings keys
   • (Backend) DataLoaders in src/dataLoaders/, feature flag infrastructure in src/domain/config/, event bus usage, Lambda functions in src/functions/

Phase 3: Plan the Implementation

7. Break the work into discrete, independently deliverable tasks. Each task must:
   • Have a single clear concern
   • Be completable in one PR or less
   • Follow natural dependency order
8. Dependency ordering:
   • iOS: Data layer → domain → presentation (models → repository → view model → view)
   • Backend: Schema/model + migrations → repository → service → API resolver → event listeners → cleanup
9. Per-task detail to include:
   • Exact file paths to create or modify
   • Existing patterns or utilities to reuse
   • Whether unit tests need to be added or updated
   • (Backend) Whether a DataLoader needs to be created or extended
10. Estimate each task using the platform-specific story point scale.
11. Detect feature flag creation. Scan the plan for any task that introduces a new feature flag (A/B test, experiment, gradual rollout, kill switch). For each new flag, generate a paired cleanup task per Feature Flag Cleanup below. The cleanup task targets the next sprint, not the current one.

Phase 4: Confirm with User

12. Present the plan as a table. Add a Sprint column so the user can see at a glance which tickets land in the current sprint vs. the next one (feature flag cleanup tickets always show next):

| # | Task | SP | Sprint  | Files affected |
|---|------|----|---------|----------------|
| 1 | ...  | 2  | current | path/to/file.swift |
| 2 | Remove <flag> feature flag and keep <variant> flow | 1 | next | path/to/file.swift |

13. Include:
    • Total story points (broken out by sprint if any tickets target the next sprint)
    • Any open questions (ambiguous scope, missing specs, unclear domain ownership)
    • Any tasks folded together and why
    • For each feature flag cleanup ticket: which experiment flow is being kept and whether the user needs to confirm it
14. Ask for confirmation before creating any Jira tickets. Confirm both the current-sprint plan and the next-sprint cleanup tickets.

Phase 5: Create Jira Tickets

15. Once confirmed, create each Task in the <JIRA_KEY> project via the Atlassian MCP:
    • Summary: [FEATURE] <concise task description>
    • Description: What to implement, which files to touch, which patterns to follow, key technical notes (include test requirements)
    • Labels: Per platform configuration above
    • Assignee: As specified by the user
    • Additional fields: {"customfield_10034": <points>, "parent": "<EPIC-KEY>", "customfield_10020": <sprint_id>}
    • Sprint targeting:
      • Implementation tickets → current sprint ID (state active)
      • Feature flag cleanup tickets → next sprint ID (state future, first result)
      • If no future sprint exists, follow the fallback set in the Jira Configuration section
16. Present the final ticket summary, grouped by sprint:

| # | Task | SP | Sprint  | Key |
|---|------|----|---------|-----|
| 1 | ...  | 2  | current | <JIRA_KEY>-XXXX |
| 2 | ...  | 3  | current | <JIRA_KEY>-XXXX |
| 3 | Remove <flag> feature flag and keep <variant> flow | 1 | next    | <JIRA_KEY>-XXXX |
|   | **Total (current)** | **5** | | |
|   | **Total (next)**    | **1** | | |

---

Architecture Reference (template — adapt to your project)

Replace the sections below with the structure of your own project. The default template assumes a Tuist-based iOS app + a Node.js GraphQL backend; trim or rewrite the sections that don't apply.

iOS (Swift / SwiftUI) — example layout

Projects/
  App/<scheme>/         → Main app target
    Modules/           → Feature modules
  UIComponents/        → Reusable UI components (design system)
  SharedModels/        → Shared data models across modules
  Service/             → Network layer, repositories
  FeatureToggle/       → Feature flags
  Analytics/           → Tracking events
  AppStrings/          → Localization

Architecture: MVVM + Coordinator + Factory (adapt to your stack)

Data pipeline: GraphQL query → Response model → Mapper → Repository → ViewModel → View

Key conventions (customize per project):

• Fonts: <font-scale-tokens> tokens only — no Font.custom(...)
• Colors: design-system color tokens only — no hardcoded hex
• Strings: localization module only — no hardcoded user-facing strings
• Tests: mandatory for every feature

Backend (Node.js / TypeScript) — example layout

src/api/              → GraphQL typeDefs + resolvers + validations + permissions (thin layer)
src/domain/           → Business logic grouped by bounded context (preferred for new code)
  <context-1>/        → e.g. user, billing, content — one folder per domain
  <context-2>/
  config/             → Feature flags
src/services/         → Legacy modules (do not add new code here)
src/functions/        → Serverless handlers (Lambda, Cloud Functions, etc.)
src/integrations/     → External APIs
src/dataLoaders/      → Per-request batched loaders (prevent N+1)

Architecture: Domain-driven layers (API → Service → Repository → Model)

Key conventions (customize per project):

• New features in src/domain/, never src/services/
• Resolvers stay thin — business logic in services
• Schema changes need migrations
• Use an event bus for cross-domain side effects
• DataLoaders for any relationship lookup

---

Guidelines (Both Platforms)

• Reuse over create: Always look for existing models, components, patterns before proposing new ones
• Data layer first: Order tasks so data/model changes come before UI/API changes
• One concern per task: Don't mix layers in the same task
• Rename/cleanup last: Renames and dead code cleanup should be the final tasks
• Be specific: Task descriptions should mention exact file paths and what changes
• Don't over-split: 3–7 tasks is ideal. Don't create a task for trivial one-line changes
• Fold trivial work: Feature flags, env vars, config values go into the adjacent task, never standalone
• Pair every new feature flag with a cleanup ticket in the next sprint: see Feature Flag Cleanup

---

Feature Flag Cleanup (Both Platforms)

Feature flags are introduced to ship safely, run experiments, or roll out gradually — but a flag left in the code after the experiment ends turns into dead code that nobody owns. To prevent that, every time a task creates a new flag this skill must also create a paired cleanup ticket scheduled for the next sprint.

When this rule fires

A task triggers a cleanup ticket if it introduces any of the following:

• A new A/B test or experiment flag
• A new gradual rollout / kill switch
• Any new entry in the feature flag system (iOS FeatureToggle module, Backend src/domain/config/)
• Any conditional that branches on a remote config value not previously used

If the task only reads an existing flag, no cleanup ticket is needed.

The cleanup ticket

For each new flag, generate one cleanup task with the following shape:

• Summary: [FEATURE] Remove <flag_name> feature flag and keep <kept_variant> flow
  • <kept_variant> is the experiment branch the team expects to win (e.g., treatment, control, variant_A, the new UI, etc.)
  • If the winner is not known at planning time, use winning variant and call out the TBD in the description
• Description must include:
  1. The flag identifier and where it is defined
  2. A short description of both flows (control vs. treatment / off vs. on)
  3. Which flow to keep (or "Winning variant TBD — update before pickup" if unknown)
  4. The files that read the flag (from Phase 2 exploration) plus the dead-flow files that will be deleted
  5. A reminder to remove the flag entry from the feature flag system itself
  6. A note that tests asserting the dead flow must be removed and tests for the kept flow must remain green
  7. (iOS) Reference the /feature-flag-remove skill as the execution playbook
• Story points: Start at 1 (mechanical cleanup). If the flag is read in many places or branches across multiple modules, split into multiple cleanup tickets rather than going above 3 SP.
• Labels: Same as the creation ticket's platform labels, plus report-tech-improvement (iOS) since this is internal cleanup, not user-facing work.
• Assignee: Same as the creation ticket unless the user specifies otherwise.
• Epic: Same epic as the creation ticket.
• Sprint: Next sprint (state future, first result from board 105). See Jira Configuration for the fallback when no future sprint exists.

What to ask the user

Before creating the cleanup ticket, confirm:

1. Which experiment flow should be kept? (treatment / control / specific variant)
   • If the experiment outcome is genuinely unknown, accept "TBD" and embed that in the ticket description.
2. Is the next sprint the right target? (default yes — only ask if no future sprint exists or the user has previously specified a different cadence)

Do not ask about whether the cleanup ticket should exist — it is mandatory.

Anti-patterns

• Don't create a standalone "Add the feature flag" ticket in the current sprint. The flag setup folds into the implementation ticket (trivial task rule).
• Don't put the cleanup ticket in the current sprint — the whole point is that the cleanup happens after the experiment runs.
• Don't skip the cleanup ticket because "we'll remember" — institutional memory is exactly what this rule exists to replace.
• Don't combine cleanup of multiple unrelated flags into one ticket. One flag, one cleanup ticket.

---

Example Output

iOS Example — "Add feed shortcuts to editorial feed"

#	Task	SP	Key
1	Implement new editorialFeedItemsWithShortcuts query and response models	3	<JIRA_KEY>-XXXX
2	Change feed container from HStack to horizontal ScrollView	1	<JIRA_KEY>-XXXX
3	Decode new response into dynamic FeedShortcuts buttons	3	<JIRA_KEY>-XXXX
4	Rename CategoryButton to FeedShortcutsButton	1	<JIRA_KEY>-XXXX
5	Remove hardcoded NEW tag code	2	<JIRA_KEY>-XXXX
	Total	10

Backend Example — "Add travel history export endpoint"

#	Task	SP	Key
1	Add exportedAt field to UserTravel Mongoose model + migration	1	<JIRA_KEY>-XXXX
2	Add findExportableByUserId to userTravel.repository.ts	2	<JIRA_KEY>-XXXX
3	Implement exportTravelHistory in userTravel.service.ts with CSV serialization + unit tests	3	<JIRA_KEY>-XXXX
4	Add exportTravelHistory mutation to user-travel-api.ts (typeDefs, resolver, Joi, shield)	2	<JIRA_KEY>-XXXX
5	Remove legacy travelDump debug resolver	1	<JIRA_KEY>-XXXX
	Total	9

iOS Example with Feature Flag — "A/B test new editorial feed card layout"

The implementation introduces a editorial_card_v2_enabled flag. The flag setup itself is folded into task 2 (no standalone "add flag" ticket), and a paired cleanup ticket lands in the next sprint.

#	Task	SP	Sprint	Key
1	Build EditorialFeedCardV2 component from Figma	3	current	<JIRA_KEY>-XXXX
2	Gate EditorialFeedItem behind editorial_card_v2_enabled flag (add flag entry + branch rendering)	2	current	<JIRA_KEY>-XXXX
3	Add analytics events for V2 card impression/tap	2	current	<JIRA_KEY>-XXXX
4	Remove editorial_card_v2_enabled flag and keep V2 flow (delete legacy EditorialFeedCard, remove flag entry, drop dead-flow tests)	1	next	<JIRA_KEY>-XXXX
	Total (current)	7
	Total (next)	1

---

Source: carloshpdoc/ios-workflow-claude — distributed by TomeVault.