Using Skills in Chat

Creating a skill is only half the value — the other half is loading it into the chat where you’re working. When a skill is loaded, the agent sees its instructions, its example plan, and its description as part of its working context, and it weights its decisions accordingly.

The Skill Loader

Every chat panel has a skill loader button (a graduation-cap icon, the same one that marks the Skills section in the sidebar). Click it to open the loader, which shows every skill you can see — your personal skills and any your admins have shared org-wide.

From the loader you can:

• Search — filter the list by typing into the search box. Searches both titles and descriptions.
• Load a skill — click any skill to add it to the current chat. It joins the loaded skills list at the top of the chat panel, marked with a small pill.
• Unload a skill — click the × on the pill to remove it from the chat. The agent immediately stops using it.

You can load skills mid-conversation. The next message you send will incorporate the newly-loaded skill’s guidance.

The Five-Skill Cap

A single chat can have at most five skills loaded at once. This is intentional: every loaded skill adds to the agent’s working context, and beyond about five the guidance starts to dilute itself. Two skills with conflicting advice will produce muddled output; five skills with overlapping concerns crowd out the user’s actual question.

If you hit the cap, the loader shows a counter (“5 / 5 loaded”) and prevents further additions until you unload one.

Curate Per Conversation

You don’t need to load every relevant skill at once. Pick the two or three that match the current question and unload them when the conversation pivots. The cost of toggling is low; the benefit of focused context is high.

How the Agent Uses Loaded Skills

Loaded skills don’t auto-execute. The agent reads their instructions, example plans, and code examples and uses them as guidance when it’s deciding what to do — choosing data sources, picking tools, ordering steps, naming output columns.

You’ll see the influence in two ways:

• Plan steps mirror the example plan. When a loaded skill has an example plan that fits the question, the agent typically reuses the same step shape, even if the columns or filters change.
• Language and naming match the instructions. If your skill says “always call this metric Net New ARR,” the agent will use that name even if the user asked for “annual recurring revenue.”

The agent still adapts to the user’s actual question. If a loaded skill is about monthly rollups but the user asks for a weekly breakdown, the agent will reuse the structure but change the time grain. Skills are a strong hint, not a script.

Multiple Skills in One Chat

When two or more skills are loaded, the agent reconciles their guidance. In practice this works best when the skills are complementary rather than overlapping — a “monthly revenue rollup” skill paired with a “Q1/Q2/Q3/Q4 rollup” skill might confuse the agent because they answer the same question at different grains.

Pairings that tend to work well:

• A data-shape skill plus an analysis skill. “Customer health score formula” + “Quarterly cohort retention” — the first defines a metric, the second uses it.
• A constraint skill plus a topic skill. “Always exclude internal email domains” + “Top 10 customers by usage” — the constraint applies broadly; the topic skill defines the question.

Pairings that tend to conflict:

• Two skills that compute the same metric differently. Pick one.
• Skills with contradictory naming conventions. Standardize before sharing org-wide.

Suggestions on the Home Screen

The home screen surfaces skill suggestions when you have skills available — both your own and those shared with your org. These suggestions are pre-filled chat prompts that load the relevant skill automatically, so you can pick them up with one click. If you’d rather not see suggestions for a particular skill, you can edit its description to be more specific about when it applies — the suggestion engine uses the description to decide when to surface a skill.

Loading Skills in Embedded Chats

If you’re using Querri inside an embedded SDK (a customer-facing chat embed), the host application controls whether skill loading is available. Embed sessions can be configured to pre-load specific skills without showing the loader UI, or to hide skills entirely. See your embed configuration for details.

When to Unload a Skill

Unload a skill when:

• The conversation has moved past the topic the skill covers.
• The skill’s guidance is fighting with what you actually want to do (it happens — skills are biases, and biases can be wrong).
• You’re hitting the five-skill cap and want to add a more relevant one.

Unloading takes effect immediately for the next message. The skill itself is unchanged — it’s still in your library, still loadable in any future chat.