Foundations for Builders: Getting Started with Claude Code
A beginner, hands-on introduction to Claude Code — the agentic loop, the feature ladder from chat to agent teams, CLAUDE.md, Plan Mode, Skills, and MCP — plus the habits that make it useful day to day.
Events
| Name | Organizer | Date | Location | Attendees | Links |
|---|---|---|---|---|---|
| AI Pilipinas Meetup #33: Claude Code Masterclass | AI Pilipinas (in partnership with iACADEMY) | 2026-07-18 | iACADEMY Nexus Campus, 7434 Yakal, Makati City, Metro Manila, Philippines | 70 | 📊 Slide Deck |
QR Code for this session
Abstract
Claude Code is not a chat window you paste code into — it's an agent that lives where your code lives: reading files on its own, deciding what needs to change, making the edit, and checking whether it worked. This beginner, hands-on session walks developers, researchers, and innovators through the agentic loop (Read, Decide, Act, Observe) and the feature ladder — from chat plus copy-paste, through terminal use, CLAUDE.md, Skills, Hooks and MCP, up to subagents and orchestrated agent teams — so attendees can match the right rung to the problem in front of them rather than reaching for the heaviest tool on day one.
We cover setup and a live CLI tour, how to pick the right model for a task, CLAUDE.md as the README for the AI, Plan Mode for architectural changes and unfamiliar codebases, the anatomy of a Skill (including the SKILL.md progressive-disclosure structure), the vocabulary distinction between skill, plugin, and subagent, and a fast introduction to MCP as a standard protocol for connecting Claude Code to external tools and data. The session closes with six before/after "rookie move" habit pairs — CLAUDE.md, Plan Mode, /clear, auto-accept, not pasting stack traces, and using Claude Code beyond bug-fixing — that even experienced users tend to miss.
Attendees leave with a working mental model of the tool, a checklist of setup steps, and habits they can apply in their very next coding session.
Agenda
- Opening — who's presenting, what the session covers
- Foundations — what Claude Code is, the agentic loop (Read, Decide, Act, Observe), who it's for
- Setup — live install, a quick CLI tour, picking the right model for the job (Fable, Opus, Sonnet, Haiku)
- The Primitive + Maturity Ladder — the six-step loop behind every task, and the seven-rung feature ladder from chat plus copy-paste to agent teams
- Why It Matters — how Anthropic uses Claude Code internally, the workflow shift it enables, how it augments the SDLC, and where it runs
- CLAUDE.md + Plan Mode — teaching Claude your repo once; reviewing an approach before any files are touched
- Skills — anatomy of a Skill, prompt engineering vs. context engineering, structure of a SKILL.md file
- Vocabulary + Subagents — skill vs. plugin vs. subagent, and how subagents protect your main context
- MCP — Model Context Protocol as a standard way to connect Claude Code to external tools and data
- Rookie Moves — six before/after habit pairs for getting more out of Claude Code day to day
- Closing + Q&A
Key Takeaways
- Claude Code is an agent, not autocomplete — it reads, decides, acts, and observes in a closed loop, and you stay in review at each step
- Match the rung to the problem: chat plus copy-paste, terminal use, CLAUDE.md, Skills and slash commands, Hooks and MCP, subagents and plugins, and agent teams are different tools for different-sized problems, not a ladder you must climb
- CLAUDE.md is the highest-leverage habit — five minutes once means every future session starts pre-briefed instead of re-explaining the repo
- Plan Mode pays off most on the changes where mistakes are costly to undo — it front-loads review before any files are touched
- A Skill is know-how (instructions); a subagent is a worker with its own context window — a subagent can use a skill, but a skill doesn't spawn anything on its own
- MCP is a standard protocol for connecting Claude Code to external tools and data sources, running under your own credentials and audit surface
- Model choice matters: Sonnet for most day-to-day work, Opus for gnarly refactors or unfamiliar codebases, Haiku for cheap and fast background work, Fable for multi-hour autonomous tasks