AI Skills are specialized modes that change how the Catalio assistant responds. Each skill gives the assistant a different area of expertise.
In AI Chat you talk to one of five agents, and each agent always answers with its own skill. Elsewhere, Catalio picks the skill from what you wrote.
How Skill Selection Works
In AI Chat: the agent’s own skill
Every message in an agent’s conversation is answered with that agent’s skill, whatever you write:
| Agent | Skill |
|---|---|
| Product Manager | Product Manager |
| Architect | Technical Architect |
| QA Analyst | QA / Testing |
| Security Analyst | Cyber Security |
| DevOps | DevOps |
Trigger words play no part here. Asking Product Manager about a security policy gets a Product Manager answer. To get a different kind of expertise, switch to another agent’s conversation.
Everywhere else: picked from your wording
Where there is no agent, Catalio looks at your wording and picks the skill whose area of expertise best fits. This covers AI clients that ask Catalio through its MCP server without naming a skill. If a specialized skill matches, you get that expertise. If nothing specific fits, the General Assistant handles the request.
Each skill has a fixed priority, and Catalio checks skills in that order, activating the first one whose trigger words match your message. General Assistant is always checked last, so it only takes over when nothing more specific matches. For example, “run a threat model on the login flow” matches Cyber Security’s trigger words and activates it before General Assistant is ever considered. This is invisible to you; you just see the AI responding in the most appropriate style.
The Conversational Skills, plus the Requirement Grader
Where skills are picked from your wording, six conversational skills can be chosen from trigger words. The Software Engineer and DevOps skills are never picked that way: Software Engineer answers only in your existing Software Engineer conversations, and DevOps only in DevOps conversations. QA / Testing and Cyber Security still match keywords outside an agent conversation, and they are also the fixed skills of the QA Analyst and Security Analyst agents. The Requirement Grader is different: it’s an operation-role skill that primarily powers the grading feature, but it also has its own high-priority chat triggers (phrases like “grade this requirement” or “assess requirement quality”), so a direct grading request routes to it too wherever skills are picked from your wording.
General Assistant
The default skill used when no other skill fits. Helps with general questions, listing and searching records, navigation, and handing off to a more specialized skill when one becomes appropriate. Also surfaces interactive blocks when a structured choice is needed.
Product Manager
The fixed skill of the Product Manager agent. Where skills are picked from your wording, it also activates when you mention: user story, requirement, acceptance criteria, process, workflow, feature request, sprint, and similar phrases.
Specializes in capturing requirements through natural conversation, writing grade-ready user_want and user_benefit fields, drafting use cases in Gherkin format, and helping with persona identification. When editing existing records, it follows a strict propose-then-review flow: it drafts changes, sends you to the form, and waits for you to confirm before anything is saved.
Technical Architect
The fixed skill of the Architect agent. Where skills are picked from your wording, it also activates when you mention: architecture, API, database, schema, integration, capability, system design, technical decision, technical architect, infrastructure, scalability, performance, microservice, system component.
Specializes in system architecture, capability mapping, API design, technical policies, and data modeling. Presents recommendations with explicit tradeoffs before documenting anything.
Software Engineer
The fixed skill of the Software Engineer agent, which is no longer offered for new conversations; it still answers in your existing Software Engineer conversations. It has no trigger words. It turns a development request into one task on your sandbox, follows that task through to done, and answers implementation questions about the code checked out there.
QA / Testing
The fixed skill of the QA Analyst agent. Where skills are picked from your wording, it also activates when you mention: test, testing, QA, quality, coverage, edge case, acceptance test, test case, bug, defect, regression, scenario, validation, verification.
Specializes in test case design, edge case identification, test strategy, and acceptance criteria verification.
Cyber Security
The fixed skill of the Security Analyst agent. Where skills are picked from your wording, it also activates when you mention: security, vulnerability, compliance, encryption, authentication, authorization, threat, policy, risk, SOC2, GDPR, HIPAA, PCI, penetration, audit, access control.
Specializes in security requirements, compliance frameworks, threat modeling, and security-control documentation.
DevOps
The fixed skill of the DevOps agent, and used nowhere else. It has no trigger words. It works through deployments, incidents, and infrastructure configuration with you, and can start a task on your sandbox when the fix is a code change.
Initiative Management
Activates when you mention: initiative, strategic intent, scope summary, process domain, create an initiative, or phrases like connect X to initiative.
Specializes in creating and managing Initiatives. Can set name, description, strategic intent, scope summary, process domains, start date, and target date. Also handles connecting entities (requirements, personas, journeys, policies, and more) into an Initiative’s scope.
Requirement Grader (operation)
Unlike the conversational skills above, the Requirement Grader is an operation-role skill — it primarily runs as part of the grading feature rather than being picked by general conversational wording. It does, however, have its own chat triggers (“grade this requirement,” “assess requirement quality,” and similar phrasing) checked ahead of the conversational skills, so asking for a grade directly activates it wherever skills are picked from your wording. In an agent’s conversation the agent’s own skill answers instead.
It evaluates a requirement across Clarity, Completeness, Feasibility, and Quality — the same rubric the AI Grader uses in the background. This is what produces the interactive explanation of a grade and targeted advice on improving a specific dimension.
What You See
In AI Chat, the agent you picked is the skill you get, and its name and mark are on every reply. A question about acceptance criteria in Product Manager’s conversation produces a structured draft with Gherkin use cases. A question about API design in Architect’s conversation produces a recommendation with tradeoffs.
Where skills are picked from your wording, the choice is invisible: the only signal is the response’s behavior. If it seems off, steer it explicitly. “Help me write a security policy for data retention” activates Cyber Security even if your original phrasing did not include a trigger word.
Skills and the AI Grader
The Requirement Grader skill and the AI Grader background feature share the same rubric. Whether the Requirement Grader skill grades a requirement on request, or the background grader runs automatically after an edit, the criteria are identical, so interactive advice and saved grades always agree.
Skills and AI Chat Tools
Skills change the AI’s expertise and communication style, not what it can do. In an agent conversation, the agent’s role decides which records it can change. Every agent can still read everything. Skills govern how the AI thinks; tools determine what the AI can act on.
Best Practices
Pick the agent for the job. In AI Chat the agent decides the skill: Product Manager for requirements and priorities, Architect for designs and data models, QA Analyst for tests and coverage, Security Analyst for security review, and DevOps for deployments and incidents. Any of them can run work on your sandbox.
Ask from the page you are working on. Each agent knows which page you asked from, so on an Initiative’s page “this initiative” needs no further explanation.
Write naturally elsewhere. Where skills are picked from your wording, trigger words are designed around how people actually speak. You do not need to memorize phrases. Being specific about the domain (“help me think through test coverage for the payment flow”) picks the right skill and produces a better response.
Relationships at a Glance
| Entity | Relationship |
|---|---|
| AI Chat | Each agent answers every message in its conversations with its own skill |
| AI Grader | Shares the Requirement Grader skill’s rubric for automatic background grading |
| Requirements | Primary artifact for Product Manager and Requirement Grader skills |
| Initiatives | Primary artifact for the Initiative Management skill |
Next Steps
- See AI Chat for the five agents, their conversations, and what they can do
- Read AI Grader to understand how the grading rubric works in detail
- Explore Initiatives to understand the scope that the Initiative skill manages
Support
If skill selection seems wrong, try restating your question with a clearer domain signal. If the AI’s response style still does not match what you need, contact support@catalio.ai.
