Agent Skills: Hamburger Method - Vertical Story Slicing

Hamburger Method - Vertical Story Slicing

Expert in applying the Hamburger Method (by Gojko Adzic) to break down large features into small, safe, deliverable vertical slices.

When to Use This Skill

Use when:

• Feature feels large but not obviously splittable with story-splitting heuristics
• User asks "how to slice" or "how to deliver incrementally"
• Need to generate multiple implementation options
• Want to compose end-to-end vertical slices

Do NOT use when:

• Story has obvious "and", "or", "manage" indicators (use story-splitting instead)
• User is asking HOW to implement (use micro-steps-coach instead)
• Feature is already small (< 1 day work)

Core Process

When user describes a feature or story that needs slicing:

1. Identify Layers (Technical or Logical Steps)

Identify the main technical or business steps involved.

List 3-6 layers that form the complete flow.

Example for notification system:

• Layer 1: Detect triggering event
• Layer 2: Decide whom to notify
• Layer 3: Format the message
• Layer 4: Deliver the message
• Layer 5: Record delivery status

2. Generate 4-5 Options per Layer

This is MANDATORY: For EACH layer, generate at least 4-5 implementation options, from simplest to most complete.

Use a numbered system: 1.1, 1.2, 1.3... for Layer 1, then 2.1, 2.2, 2.3... for Layer 2, etc.

Quality gradient (low to high):

• Manual / hardcoded
• Semi-automated / configurable
• Fully automated / robust
• Scalable / multi-channel
• Enterprise-grade / resilient

Example for "Deliver the message" layer:

• 4.1: Manual email from your personal account
• 4.2: Scripted email via command line
• 4.3: Email via SMTP service (no retries)
• 4.4: Email via queuing system with retries
• 4.5: Multi-channel (email, push, SMS) with fallbacks

3. Force Radical Slicing

Always ask this question out loud in your response — do not skip it:

"If you had to ship something by tomorrow, what would you build?"

Then answer it explicitly by naming the specific options (by number) you would pick from each layer. This forces the absolute minimum viable slice, not just a "simple version".

Quick Reference: Quality Gradients

Use this table to systematically generate 4-5 options per layer from simplest to most complete.

| Layer Type | Level 1 (Manual) | Level 2 (Scripted) | Level 3 (Automated) | Level 4 (Scalable) | Level 5 (Enterprise) | |------------|------------------|-------------------|---------------------|--------------------|--------------------| | Trigger/Detection | Manual check | Scheduled script | Event-driven | Real-time stream | ML-powered | | Data Source | Hardcoded | Single file/DB | Multiple sources | External APIs | Federated | | Processing | Manual steps | Script | Background job | Queue system | Distributed | | Validation | None | Basic checks | Business rules | Comprehensive | Anomaly detection | | Output/Delivery | Manual action | Email/file | API call | Multi-channel | Personalized | | Monitoring | None | Console log | DB record | Dashboard | Real-time alerts |

Example usage:

• Layer: "Deliver notification"
• Quality dimension: "Output/Delivery"
• Options: (1) Manual email, (2) Scripted email, (3) SMTP service, (4) Queue + multi-channel, (5) Personalized dynamic content

4. Filter & Prioritize Options

Eliminate options that:

• Are too costly for the value they provide
• Block fast delivery
• Are irreversible

Keep options that are:

• Fast to build (ideally hours, max 1-3 days)
• Testable with real users
• Reversible or expandable later

5. Compose a Vertical Slice

Help the user select ONE option from EACH layer to form a complete end-to-end slice.

If the user requested multiple outputs or metrics (e.g. "revenue, top products, and trends"), the first slice must pick EXACTLY ONE — the most informative single output. Delivering all of them in slice-1 defeats the purpose of incremental validation.

Criteria for a good slice:

• Delivers value to at least one real user or stakeholder — name them explicitly
• Names the specific decision or question the slice answers for that person
• Can be tested in production (or near-production)
• Takes less than 1-3 days to build
• Preserves system stability (zero downtime)

6. Plan Next Slices

Once the first slice is identified, suggest 2-3 follow-up slices that:

• Improve one layer (e.g., automate what was manual)
• Add a missing quality attribute (e.g., logging, error handling)
• Expand reach (e.g., more users, more scenarios)

Rules of Thumb

• Every vertical slice must be usable by someone — even if it's just one test user or internal stakeholder
• You don't need the "best" version first — just the smallest that teaches you something
• Always list options explicitly — don't just describe "simple vs. complex"; give concrete examples
• Push for the lowest quality that still works — hardcoded values, manual steps, no error handling if it helps ship faster

Example Interaction

User: "We need to notify users when a product they're watching drops in price"

You respond:

Let's apply the Hamburger Method to slice this feature.

Layers identified:

1. Detect price change
2. Identify watching users
3. Format notification
4. Deliver notification
5. Track delivery

Options per layer:

Layer 1 - Detect price change:

• 1.1: Manual check once per day
• 1.2: Cron job checking specific products
• 1.3: Automated price scraping for all products
• 1.4: Real-time event-driven detection
• 1.5: ML-based anomaly detection for price drops

Layer 2 - Identify watching users:

• 2.1: Hardcode one test user
• 2.2: Query existing watchlist table
• 2.3: Multi-tier watchlist with preferences
• 2.4: User segmentation based on behavior
• 2.5: Personalized relevance scoring

Layer 3 - Format notification:

• 3.1: Plain text string
• 3.2: Simple template with product name + price
• 3.3: HTML email with branding
• 3.4: Rich notification with images and CTAs
• 3.5: Personalized dynamic content

Layer 4 - Deliver notification:

• 4.1: Manual email from personal account
• 4.2: Scripted email via Gmail API
• 4.3: SMTP service (no retries)
• 4.4: Email queue with retries
• 4.5: Multi-channel (email + push + SMS)

Layer 5 - Track delivery:

• 5.1: No tracking
• 5.2: Log to console
• 5.3: Store delivery status in DB
• 5.4: Dashboard with delivery analytics
• 5.5: Real-time monitoring with alerts

Smallest vertical slice (ship by tomorrow):

• 1.1: Manual price check
• 2.1: Notify one test user (you)
• 3.1: Plain text message
• 4.1: Send via personal email
• 5.1: No tracking

This slice can be deployed today to validate the concept with zero infrastructure.

Next slices:

• Slice 2: Automate price detection (1.2), keep rest the same
• Slice 3: Expand to real watchlist users (2.2)
• Slice 4: Add basic SMTP delivery (4.3)

Reference

For full details on the Hamburger Method, see hamburger-method.md in this skill directory.

Author: Gojko Adzic Source: https://gojko.net/2012/01/23/splitting-user-stories-the-hamburger-method/

Coaching Tone

• Be pushy about generating ALL options (don't skip this step)
• Challenge the user if they propose a slice that's too big
• Always ask: "Can we make it even smaller?"
• Use Eduardo Ferro's phrases: "What if we only had half the time?" "Can we avoid doing it?"

Integration with Other Skills

This skill works in sequence with other skills:

Typical workflow:

1. story-splitting: Detect and split oversized stories with obvious red flags
2. hamburger-method (THIS SKILL): For stories that are large but not obviously splittable, generate layers + options
3. complexity-review: Review proposed vertical slice, simplify if needed
4. micro-steps-coach: Break chosen vertical slice into 1-3h implementation steps

Use this skill when:

• Feature is large but doesn't have obvious "and", "or", "manage" indicators
• Need to explore multiple implementation options systematically
• Want to compose minimal end-to-end slice

Vs. story-splitting:

• story-splitting: Best for stories with clear linguistic red flags ("manage users and roles")
• hamburger-method: Best for features that need layered analysis ("implement notifications")
• Can use BOTH: Split story first, then apply hamburger method to each smaller story

Integration example:

• User: "Implement user notifications" (no obvious split points)
• Apply hamburger-method → Identify 5 layers, generate options, compose smallest slice
• Then use complexity-review → Ensure simplest slice is truly simple
• Then use micro-steps-coach → Break slice into 1-3h steps

Self-Check: Did I Apply This Correctly?

After applying this skill, verify:

• [ ] I identified 3-6 clear layers (not too many, not too few)
• [ ] I generated at least 4-5 options per layer (not just "simple vs. complex")
• [ ] Options follow a quality gradient (manual → scripted → automated → scalable → enterprise)
• [ ] I forced radical slicing by asking "ship by tomorrow"
• [ ] The smallest vertical slice uses level 1-2 options from each layer
• [ ] The smallest slice delivers value to at least one user (even if it's just me)
• [ ] The smallest slice can be deployed in less than 1-3 days
• [ ] I proposed 2-3 follow-up slices showing incremental improvement

If any checkbox fails, revisit the process.

Red flags that I didn't do this right:

• Layers are too technical ("frontend, backend, database") instead of functional
• Only 2 options per layer ("manual or automated")
• Smallest slice still requires new infrastructure (Redis, Kafka, etc.)
• Smallest slice would take more than 3 days to build
• Follow-up slices aren't clear improvements over previous slices