Claude Code

Anthropic's agentic coding tool: an agent (see Agents) that lives in the terminal, reads and edits files, runs commands, and drives a task to completion. Built on Claude. It is what powers this vault's read/write integration. See Architecture/Knowledge Vault/Knowledge Vault.

Also runs as a desktop app, a web app at claude.ai/code, and IDE extensions for VS Code and JetBrains. The terminal is the canonical form.

How it works

  • Runs the agent loop directly against your filesystem and shell: observe the repo, plan, edit, run, repeat.
  • Tool calls go through a permission layer so it asks before doing anything destructive or outward-facing.
  • Holds a large context and compacts the conversation as it grows, so long sessions survive.

Extension points

These are how you bend it to a workflow, and how the vault integration is built:

  • Skills: triggered .md capability files. See Agent Skills and Agent skills I use.
  • Hooks: shell or prompt callbacks on lifecycle events like SessionStart and Stop. The vault uses a SessionStart hook to load daily context and a Stop hook to catch unsaved notes.
  • MCP servers: connect external tools and data through the Model Context Protocol.
  • Slash commands: user-invoked shortcuts, often backed by a skill.
  • Subagents: spawn isolated agents for parallel or fan-out work.
  • Plugins: bundle skills, hooks, and commands into one installable, version-controlled unit. The vault ships as exactly this.
  • settings.json: permissions, env vars, and hook wiring.

Notes

  • Costs follow Claude token pricing; an agent burning through tool calls adds up.
  • The plain-Markdown, local-first design of Octarine is what lets an agent touch the vault safely: every change is a readable diff that GIt can undo.