How to Use Claude Code: A Guide to Slash Commands, Agents, Skills, and Plug-ins

Teresa Torres walks through the building blocks of Claude Code: markdown context files, slash commands, agents and sub-agents, skills, hooks, and plug-ins. Her core argument is that going deep on one tool and understanding how these pieces fit together yields more value than chasing every new AI product. She explains when each block fits, why managing the context window matters, and where features like skills still fall short. The piece is practical for anyone building repeatable AI workflows, especially non-technical users, and closes by previewing paid walkthroughs of her own plug-ins.

01Key takeaways

• Commit to one AI tool and learn its building blocks deeply rather than chasing every new product.
• Keep CLAUDE.md files lean and point to reference files so Claude loads only what each task needs.
• Use slash commands when you want to decide exactly when a workflow runs; use agents when Claude should decide.
• Offload noisy searches and exploration to sub-agents to protect the main context window and speed up parallel work.
• Use hooks for steps that must run identically every time, such as injecting today's date at session start.
• Review any plug-in code before installing it, since plug-ins can include scripts that run on your computer.

02Key sections

Markdown files for context and memory

Claude Code can read and write local files, so persistent context lives in CLAUDE.md files at global and project levels. Reference files keep those memory files lean, loading only what a task needs.

Slash commands as stored procedures

Markdown prompts saved in a commands folder become invocable shortcuts, scoped globally or to a single project. They suit tasks you want to trigger explicitly and deterministically.

Agents and the context window

Claude is itself an agent, and it can spawn sub-agents or run user-defined ones. Sub-agents keep bulky search results out of the main context window and enable parallel work, at a higher token cost.

Skills and hooks

Skills bundle instructions with scripts and are portable across Claude surfaces, though Claude often fails to invoke them unprompted. Hooks run deterministic code at lifecycle events, guaranteeing behavior the LLM might otherwise vary.

Plug-ins, marketplaces and when to use each block

Plug-ins package the other blocks for sharing through git-based marketplaces, with a caution to review code before installing. A comparison table maps each block to its purpose, invoker, and portability.

03From the post

“I've been going deep on Claude Code. I'm finding this has been a better strategy for getting value out of AI—rather than trying every new tool on the market. By focusing on one tool, I've been able to learn how to use Claude Code's building”

04Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.