Track 6 — Agents, Skills & Subagents

Now you compose everything into autonomous, multi-step systems. Agent Skills package reusable expertise Claude can invoke on demand; subagents let a main agent delegate focused work to specialized helpers with their own context; and orchestration coordinates several agents on one goal. This is the heart of the "agentic" material — and the largest single domain on the CCA-F exam.

Audience: developers. · Est. time: ~8–10 hours over a week. · Prereq: track-5-mcp-and-tools (tools/MCP) and track-4-claude-code. · Outcome: you can build skills, subagents, and multi-agent systems.

Sequence

1. Agent Skills · ⏱ ~2.5–3 h + project

• Course: introduction-to-agent-skills — what a skill is (a packaged, model-invoked capability with instructions and optional tools/resources), when to write one vs. a plain prompt, anatomy and authoring, and how Claude decides to use a skill.
• Project: p13-agent-skill — author a skill that solves a recurring task of your own, then trigger it from Claude.

2. Subagents · ⏱ ~3 h + project

• Course: introduction-to-subagents — why delegate to subagents (separate context windows, focused roles, parallelism), defining them (e.g., an Explore/research subagent), the main-agent-vs-subagent boundary, and the agentic patterns from "Building Effective Agents" (prompt chaining, routing, orchestrator-workers, evaluator-optimizer).
• Project: p14-subagent-research — define a research subagent and have a main agent delegate a multi-source investigation to it, then synthesize the result.

3. Multi-agent orchestration · ⏱ ~3 h + project

• Project: p15-multi-agent-orchestrator — build an orchestrator that fans work out to multiple worker agents (at least one backed by an MCP tool from track-5-mcp-and-tools) and merges their outputs. Watch context, cost, and failure handling.

You're ready to move on when…

• You can decide when a task warrants a skill, a subagent, or just a better prompt.
• You've authored a working skill and a working subagent.
• You can name the core agentic patterns (prompt chaining, routing, orchestrator-workers, evaluator-optimizer) and pick the right one for a problem.
• You've built a multi-agent orchestrator and can discuss its context/cost/failure trade-offs.
• You've completed p13, p14, and p15.

See also

• README — the full track map.
• Next: track-7-deployment-and-scale — deploy on a cloud with governance.
• Backing 301 note: cca-domain-1-agents (the highest-weighted exam domain).
• progress — tick these off.