Using Civic Skills
A Skill is a set of instructions, business context, terminology, and workflows that the AI loads on top of its current toolkit. Skills teach the agent how to use tools — not which tools are available.
Skills vs Toolkits
Skills and Toolkits work together but serve different purposes:
| Toolkit | Skill | |
|---|---|---|
| What it is | A set of tools (MCP servers) | A set of instructions and business context |
| How it works | Switch to it | Load on top of current toolkit |
| Purpose | Defines what tools are available | Defines how to use them |
| Replaces current context? | Yes | No — layers on top |
| Multiple active at once? | No | Yes |
| Example | "Support" toolkit with Gmail + Slack | "Escalation Procedures" skill |
A toolkit switch replaces the entire tool context. Loading a skill adds domain knowledge and workflow instructions without removing any tools.
How to create and manage toolkits
Use Cases
PR Reviewer — Load your code review standards, naming conventions, and style guide. The agent knows how you review PRs before it looks at a single diff.
Pipeline Summary — Load your Salesforce stage definitions, deal naming conventions, and reporting format. The agent produces consistent summaries without re-explanation each session.
Support Agent — Load your escalation policies, SLA thresholds, and tone guidelines. The agent handles routine cases and escalates correctly without needing to re-read the playbook each time.
Onboarding Companion — Load your new hire checklist, internal tool guides, and team contacts. The agent walks new team members through setup consistently.
How to Create a Skill
- Log in to app.civic.com
- Click Create Profile (or "Create Toolkit")
- Set the profile type to SKILL
- Give it an alias (e.g.
pr-reviewer,escalation-procedures) - Add your instructions in the content field
Writing effective skill instructions:
- Write in second person: "You are a code reviewer who follows these standards..."
- Include examples of correct and incorrect behavior
- Be specific about terminology unique to your business
- Keep each skill focused on one domain — multiple skills can be loaded simultaneously
How to Load a Skill
Via URL Parameter
Include the skills parameter when connecting your agent:
https://app.civic.com/hub/mcp?profile=support&skills=escalation,canned-responses
Multiple skills are comma-separated. They load in the order specified.
Via Configurator Agent in Civic Chat
"Load the escalation-procedures skill for this session"
"Add the pr-reviewer skill to the current context"
Via the load-skill Tool
If your agent has access to the load-skill tool, it can be instructed to load skills dynamically:
"Use the load-skill tool to load the pipeline-summary skill"
Multiple Active Skills
Skills layer — multiple skills can be active simultaneously without conflict. Each adds its instructions to the agent's context.
If instructions from two skills conflict, the later-loaded skill typically takes precedence, but it is better practice to keep skills non-overlapping.
Session Scope
Skills persist for the duration of the session. They are not automatically persisted to the toolkit configuration unless you explicitly save them there.
Best Practices
- One domain per skill — A skill that covers code review AND customer support AND data analysis is too broad. Split it.
- Second person voice — "You are..." establishes persona clearly and consistently.
- Include examples — "When a user asks about X, respond with Y" is more effective than abstract principles.
- Avoid instructions that duplicate guardrails — Guardrails are enforced at the protocol level and cannot be overridden. Skill instructions are guidelines that the LLM follows. Don't rely on skill instructions for security-critical constraints.