A branching skill picks which heavy reference file to load based on what you ask. You hide column-formatting and view-formatting knowledge behind one parent skill — and only pull in what the request actually needs.
Meta takeaways
Pattern
One parent skill routes to one of N reference files.
Why
Big reference files cost tokens. Don't load them every turn.
Reuse
Same pattern fits governance, policy, brand — anything heavy and conditional.
The parent skill is just a classifier
The author reworked one skill so it covers both view formatting and column formatting under a single entry point. The parent skill reads the user's request, decides which sub-area applies, and then reads the matching reference file before doing the formatting work.
The parent skill stores almost no formatting knowledge itself. Its only job is to classify the request — column, view, or both — and pull in the right reference file on demand.
This is the same shape as a router function in code. The parent owns the decision; the sub-files own the detail.
A skill isn't required to be one self-contained document. A skill can be a small dispatcher that loads other files only when needed.
Why branch instead of inlining everything
The earlier version of this skill always read a reference file on every run. That was fine because the skill always needed that knowledge. The new version covers two reference files — view formatting and column formatting — and each one is heavy. Loading both every time would waste context the model could spend on actually doing the job.
If a skill always needs a piece of knowledge, embed or always-load it. If it only sometimes needs it, branch and load it conditionally.
The author calls out the general principle: governance rules, brand definitions, policy overlays — anything you only consult sometimes — fits this same branching pattern.
Branching isn't about saving disk space or runtime. It's about keeping the model's working context focused on the knowledge that actually applies to this request.
View formatting vs column formatting
SharePoint lets you style two different things using JSON: a single column's cells, or a full list view. The PnP (Pattern and Practices) samples library publishes recipes for both. The previous video only covered view samples. This one adds the column samples and treats them as a sibling reference file behind the same parent skill.
View formatting paints rows and the surrounding chrome. Column formatting paints individual cells. The author needs separate reference files because the JSON schemas and patterns are different.
A request like add color coding to my status column obviously points at column formatting. A request like show this list as a timeline points at view formatting. The parent skill makes that call before doing any work.
The "both" path
Some requests need both reference files. The example in the video: make my list beautiful, highlight overdue rows in red, and format the priority. Highlighting overdue rows is view formatting; coloring a priority cell is column formatting. The parent skill recognises the ambiguity and loads both reference files.
A branching skill isn't strictly either-or. The classifier can pick a single branch, both branches, or — in principle — none.
Because the parent is just routing logic, adding a third branch later (say, an action-button reference file) costs you a new condition in the classifier and one new file. The existing branches don't change.
A "both" branch isn't a special third skill. It's just the classifier saying "load file A and file B" before the same downstream formatting work runs.
Flashcards — Foundations
01 · Foundations
Branching skill
tap to reveal →
A parent skill that classifies the request and loads one of several reference files on demand instead of inlining all knowledge up front.
← tap to flip back
01 · Foundations
Parent skill
tap to reveal →
The top-level skill in a branching setup. Owns routing logic only — decides which sub-file to read before doing any formatting work.
← tap to flip back
01 · Foundations
Conditional load
tap to reveal →
Pulling a reference file into the model's context only when this request actually needs it. Saves context for knowledge you sometimes need.
← tap to flip back
01 · Foundations
View formatting
tap to reveal →
SharePoint JSON that styles entire list views — row highlighting, timelines, gallery layouts. The earlier video covered this set.
← tap to flip back
01 · Foundations
Column formatting
tap to reveal →
SharePoint JSON that styles a single column's cells — colour-coded status, icons, progress bars. Different schema from view formatting.
← tap to flip back
01 · Foundations
Both-path branch
tap to reveal →
When a request spans multiple sub-areas, the classifier loads more than one reference file in the same turn instead of choosing a single branch.