Skip to main content

Skills

Skills are higher-level, goal-oriented workflows that abstract over MCP tools. They provide platform-agnostic actions that work across different implementations, using the ontology to route to the right tools automatically.

The Problem Skills Solve​

MCP tools are atomic operations: github_get_repository, gitlab_list_pipelines, circleci_get_workflow. To accomplish a goal like "check CI status," an agent needs to:

  1. Know which specific tools exist
  2. Understand which tool applies to which platform
  3. Compose multiple tool calls correctly
  4. Handle platform-specific differences

Skills solve this by providing goal-oriented actions that work regardless of the underlying platform.

How Skills Work​

You: "Check the CI status for api-service"

1. Platform looks up "api-service" in ontology
   → Entity kind: GithubRepository

2. Platform routes to skill implementation for GithubRepository
   → Uses GitHub Actions tools

3. Skill executes and returns results
   → "2 workflows passed, 1 pending"

If api-service were a GitLab project instead, the same skill would automatically route to GitLab CI tools. You don't need to know which platform—the ontology handles routing.

Skills vs Tools​

MCP ToolsSkills
Atomic operationsGoal-oriented workflows
Platform-specificPlatform-agnostic
github_list_workflowscheck_ci_status
Agent must choose correct toolPlatform routes automatically
Single API callMay compose multiple tools

Example Skills​

CI/CD Skills​

  • check_ci_status — Check pipeline status across GitHub Actions, GitLab CI, Jenkins, CircleCI, Argo CD
  • trigger_deployment — Deploy application to an environment
  • rollback_deployment — Rollback to a previous version
  • get_deployment_history — View deployment timeline

Code Quality Skills​

  • review_pull_request — Comprehensive PR review with diff analysis, CI status, and code quality checks
  • analyze_codebase_health — Aggregate code quality metrics
  • check_security_vulnerabilities — Security scanning across Snyk, Dependabot, etc.

Operations Skills​

  • check_service_health — Health checks across monitoring tools (Datadog, Grafana, Prometheus)
  • create_incident — Create incident in PagerDuty, Opsgenie, or other systems
  • notify_team — Send notifications via Slack, Teams, or email

Execution Modes​

Skills can operate in three modes:

Guided Mode​

The skill returns a contextual prompt for the AI agent to follow, along with entity context and suggested tools:

User: "Review the PR on auth-service"

Skill returns:
- Full PR context (title, description, author)
- Available tools for this entity type
- Guidance prompt for conducting the review

This mode is flexible and handles edge cases naturally.

Orchestrated Mode​

The skill directly executes MCP tool calls and returns structured results:

User: "Check CI status for api-service"

Skill executes:
1. Calls github_list_workflows
2. Calls github_get_workflow_run for each
3. Aggregates results
4. Returns: {status: "passing", workflows: [...]}

This mode is deterministic and performant—no LLM inference per step.

Hybrid Mode​

The skill gathers data via tool calls, then returns a guided prompt for analysis:

User: "Review the PR on checkout-service"

Skill executes:
1. Fetches PR details via tools
2. Fetches diff via tools
3. Fetches CI status via tools
4. Returns guided prompt with all context for agent to analyze

This combines the efficiency of orchestration with the flexibility of guidance.

Ontology-Driven Routing​

Skills leverage the ontology for intelligent routing across systems:

Skill: check_deployment_health
Entity: KubernetesDeployment "myapp-prod"

Workflow:
1. Check K8s deployment status (Kubernetes molecule)
2. Follow "deploys-from" relation → GithubRepository
3. Check CI status (GitHub molecule)
4. Follow "monitored-by" relation → GrafanaDashboard
5. Check metrics (Grafana molecule)

The skill traverses relationships in your ontology to gather context from multiple systems automatically.

Using Skills​

Skills are invoked through natural language in chat:

You: Check the CI status for api-service
AI: ✓ All checks passing
    - build: passed (2m ago)
    - test: passed (2m ago)
    - deploy-staging: passed (1m ago)

You: Review the open PR on auth-service
AI: Reviewing PR #42: "Add OAuth2 support"

    Summary: Adds OAuth2 authentication flow...

    CI Status: All checks passing

    Code Review:
    - ✓ Good test coverage
    - ⚠ Consider adding rate limiting
    - ...

Building Skills​

Skills are provided by skills molecules—specialized molecules that orchestrate other molecules' tools.

See Building Molecules for implementation details.

Next Steps​