Skills Overview

A skill is a reusable analysis recipe. You teach the AI agent how to handle a particular kind of question once, and from then on it can reach for that recipe whenever a similar question comes up — using your data, your terminology, and the steps you’ve already proven work.

Think of skills as the difference between explaining your monthly close process to a new analyst every single month versus writing it down once and handing them the playbook. The output gets more consistent, the time-to-answer drops, and the institutional knowledge stops living in one person’s head.

What a Skill Contains

Every skill bundles three kinds of guidance for the agent:

• A title and description — the human-readable name and a short summary of what the skill is for. The agent uses these to decide whether the skill is relevant to the current question.
• Advanced instructions — a free-form natural-language brief. This is where you write down the things you’d tell a teammate: which data source to use, how to interpret edge cases, what to call things, what to avoid.
• An example plan — an optional ordered list of analysis steps. Each step names the tool the agent should use (filter, group, join, transform, visualize, etc.) and the columns or dependencies it needs. The plan is a strong hint, not a hard script — the agent adapts the plan to the actual question.

You don’t have to provide all three. Instruction-only skills work well for high-level guidance (“always exclude internal email domains”). Plan-only skills work well when the steps are deterministic and the question is narrow (“monthly recurring revenue rollup”). Most useful skills combine both.

When Skills Help

Skills earn their keep when the same kind of question keeps coming back, when the analysis touches non-obvious business rules, or when consistency across people matters more than creative latitude. A few concrete patterns:

• Recurring reports — month-end close, weekly KPI rollups, board-deck financial summaries. The structure is fixed; only the time window changes.
• Domain-specific analyses — anything where the right answer depends on knowing your data. Customer-cohort retention, product-line attribution, sales-cycle stage definitions.
• Onboarding new teammates — instead of pairing on the same five-step recipe each time someone joins, share the skill org-wide and let the agent walk them through it.
• Tricky data shape — joins that need a particular bridge table, filters that have to exclude test accounts, date columns that need timezone correction.

The Easiest Way: Save from a Project’s Data Flow

By far the fastest way to build a useful skill is to capture one from a project where you’ve already done the analysis. Open the project’s Data Flow, select the steps that produced the result you liked, and click Save as Skill in the selection toolbar. The example plan is pre-filled from the actual steps you ran — no guessing required. See Creating Skills.

When Skills Don’t Help

Skills are not the right tool for every situation. Skip them when:

• The question is genuinely one-off. A skill is overhead — both to write and to maintain. If you’re not going to ask this question again, just ask it directly.
• The “right answer” changes frequently. Skills assume the recipe is stable. If your business logic shifts every quarter, you’ll spend more time updating skills than you save.
• You want the agent to explore. Skills bias the agent toward a known approach. For open-ended discovery work — “what’s interesting in this dataset?” — let it stay open.

Personal vs. Org-Shared

Every skill starts as personal — only you can see it. Admins can promote a skill to org-shared, which makes it visible to everyone in your organization. This is how skills compound: one person figures out the right way to do something, and the whole team picks it up.

Non-admins can create, edit, and delete their own personal skills freely. Sharing org-wide is gated to admins — see Sharing Skills for the full permissions model.

Where to Find Skills

Skills live in the Skills section in the left sidebar. The library shows two groups:

• My Skills — skills you created (whether or not they’re shared org-wide).
• Shared with Org — skills your admins have published organization-wide.

You’ll also see skills surfaced in the chat panel, where you can load up to five at a time to guide a single conversation. See Using Skills in Chat.

Next Steps

• Creating Skills — the two ways to author a skill, the fields, and how to write a good one.
• Using Skills in Chat — load skills into a chat and watch the agent reach for them.
• Sharing Skills — promote skills org-wide and understand the permission model.
• Importing & Exporting — move skills between organizations using the .qskill format.