agentleFS
ToddGBensonToddGBenson/keel/.claude/skills/story-splitting/SKILL.md

story-splitting

Skill0 starsChanged 3 months agoApache-2.0View on GitHub

Split an epic or oversized story into thin vertical slices that each ship independently. Use when refining an epic, when a story is too large to estimate or exceeds ~3 days, when work has been sliced by technical layer, or when asked how to break down a feature. Covers the split axes and when not to split.

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If no, it is not a slice. It may still be legitimate work — a chore, a spike, an enabler — but it is not a story and it should not be sized …

    line 39
  2. Distinctive:

    3. CRUD operation: Create, then read, then update, then delete.

    line 53
  3. Distinctive:

    If the reason the story is unsplittable is that nobody knows enough to size it, the answer is a spike: time-boxed, produces a written …

    line 92
  4. Distinctive:

    Prefer, where you can, to build the enabler *inside* the first vertical slice that needs it.

    line 104

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Story splitting

Consumers: product-owner, /refine, architect (feasibility), process/gates/g1-ready.md.

The rule

Split along user-visible value. Never along technical layers.

Layer slicing feels efficient — it matches how the work is done — and it produces a backlog where nothing is shippable until everything is finished. That defeats the entire point of incremental delivery: you lose the ability to stop early, reprioritise, or learn from something real.

✗ LAYER SLICING — nothing ships until all three land
  #1 Create the drafts table
  #2 Build the drafts API endpoint
  #3 Wire up the autosave UI

✓ VERTICAL SLICING — each ships and is independently valuable
  #1 Editor state survives an accidental tab close within the same session
  #2 Editor state survives session expiry and is offered on next sign-in
  #3 User can browse and restore any of their last 5 recovered drafts

Each vertical slice touches the database, the API, and the UI. That is correct. A slice that touches only one layer is a task, not a story.

The test

If we shipped only this one and stopped, would a user be able to do something they could not do before?

If no, it is not a slice. It may still be legitimate work — a chore, a spike, an enabler — but it is not a story and it should not be sized or prioritised as one.

The split axes

Try them in roughly this order; the first two cover most cases.

1. Workflow step. Take the user's journey and ship it a step at a time.

Checkout → ship: cart persists → ship: address entry → ship: payment → ship: confirmation email

2. Happy path vs. error path. The happy path is usually a third of the work and most of the value. Ship it, then harden.

Ship: upload succeeds → ship: upload fails gracefully with retry → ship: partial upload resume

3. CRUD operation. Create, then read, then update, then delete. Rarely all equally valuable — often delete can wait a quarter.

4. Data variation. One format, then more.

Ship: CSV import → ship: Excel import → ship: malformed-file handling

5. User role. One audience first.

Ship: individual users → ship: team admins → ship: org owners

6. Platform / interface. Web, then mobile, then API.

7. Performance as its own story. Make it work; then make it fast, with a number.

Ship: search returns correct results → ship: search p95 < 200 ms at 10k documents

This one is frequently missed. "Fast" bolted onto a functional story makes both harder to verify and hides which part is late.

8. Business rule complexity. The simple rule first, the exceptions after.

Ship: flat 10% discount → ship: tiered discounts → ship: stacking and precedence rules

9. Defer the hard operational bit. Ship behind a flag, to internal users, or to one tenant, before general availability.

When not to split

Splitting has a cost — coordination, integration, review overhead, and cognitive load on whoever holds the thread. Do not split when:

  • The pieces cannot be independently verified, so you get overhead without feedback
  • The split is purely technical and every piece is dark until the last one lands
  • The result is a story so thin its acceptance criteria are longer than its implementation
  • The seam you would create is one you will immediately have to remove

A story of ~2–3 days is fine. The ~3-day guideline is a signal to look, not a hard cap requiring a split into four half-day fragments. Fragmentation has its own failure mode: a backlog of 40 tiny items where nobody can see the shape of the feature.

Spikes are not slices

If the reason the story is unsplittable is that nobody knows enough to size it, the answer is a spike: time-boxed, produces a written answer and a recommendation, produces no production code. Code written during a spike is thrown away — keeping it is how prototypes become production systems nobody designed.

Enablers

Sometimes real infrastructure genuinely must land first. Be honest about it: label it an enabler, note which stories it unblocks, and accept that it delivers no user value on its own. What you must not do is dress it up as a user story — that hides the true cost of the feature it serves and corrupts the prioritisation data.

Prefer, where you can, to build the enabler inside the first vertical slice that needs it. The slice justifies the infrastructure and proves it works.

After splitting

Each resulting story needs its own full G1 treatment — narrative, binary acceptance criteria, non-functional criteria with numbers, relevance triage (security / AI / privacy), dependencies, estimate, test approach. See the writing-acceptance-criteria skill.

Re-check INVEST on each piece, particularly Independent. Splits frequently create hidden ordering dependencies; if #2 cannot ship without #1, say so explicitly rather than discovering it at planning.

Controls

SA-3 · SA-4(3) (development methods) · SA-15 · AI RMF MAP 2.3.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.