Analyse & Label
The skill that reads a source for you — labelled highlights, an Analysis frame of questions, assumptions, evidence, and gaps, and the graph links between them.
Analyse & Label is the fastest route from a raw document to usable evidence. It reads a source against criteria you set and comes back with three things: highlights with labels, an Analysis frame, and the links between them. Nothing is saved until you say so.
Running it
Type / in the Agent composer and pick Analyse & Label, or just type what you want analysed — analyse, label, highlights, and scan all resolve to the same skill.
- If you name a subject in the same breath — "analyse this transcript for onboarding friction" — the scan starts straight away.
- If you don't, the Agent asks what to look for first. Answer with the criteria that matter: a theme, a research question, a persona, a stage of the journey.
The scan reads the item you have open. Say what to look for before scanning — an unfocused pass produces unfocused highlights.
What comes back
Results appear under the Agent's reply as Review by type, with a tab per kind of result. Tabs only appear when there is something in them.
| Tab | What it holds |
|---|---|
| Highlights | Passages from the source, each with suggested labels |
| Frame | A summary of what the source says as a whole |
| Questions | What the source raises but doesn't answer |
| Assumptions | What is being taken as true without proof |
| Evidence | What the source actually establishes |
| Gaps | What is missing and would need more research |
Questions, assumptions, evidence, and gaps together make the Analysis frame — the same four types the research frame uses, so an analysis slots into the structure your project already has rather than sitting beside it.
Reviewing highlights
Each highlight card carries two sub-tabs:
- Content — the excerpt and its labels. Both are editable: tighten the passage so the quote is exactly what you want to cite, and edit the labels as a comma-separated list so they match the coding scheme you already use.
- Links — the graph relations the skill suggests for this highlight, such as the evidence it supports.
From there:
- Add on a card saves that one highlight.
- Add all saves everything still pending in one go.
- The × on a card dismisses a suggestion you don't want; it disappears from the review list without being saved.
Saved highlights land on the source document exactly where the excerpt sits, so they behave like highlights you made by hand — see Highlights & labelling.
Saving the Analysis frame
On any of the frame tabs, Save Analysis types writes the questions, assumptions, evidence, and gaps into the project's research frame, along with the links between them and the highlights behind them. Existing frame items are left alone — an analysis appends to what you already have rather than replacing it.
You can save the highlights, the frame, or both. Adding highlights while the frame draft is still unsaved saves both at once, and the toast tells you how many frame items were linked.
Why the links matter
The skill doesn't just produce a list. Each frame item is connected to the highlights that justify it, so an assumption in the frame is one click from the passage that suggested it, and a gap says which source it came out of. That is what keeps a synthesis auditable months later — see Change notes & history.
Next
Continue with Reviewing AI output, or see Highlights & labelling for the manual route.
Related articles
Documents: Analyse & Label