Skip to content

Knowledge base

Claude Code

Claude Code is a coding agent built as one simple loop over a small set of tools. That design leaves most of the leverage with the user: how you manage its context, how you prompt it, and how you extend it with hooks, skills, sub-agents and MCP.

Written by

Updated:

Tools 4

Our position. Whichever agent you run, its intelligence ceiling is your context management ceiling (opens in a new tab).

Its builders rely on the model rather than on scaffolding, and let Bash do the work a dedicated tool would. Because the agent gets worse as its context grows, most practitioner technique comes down to keeping that context focused (How Claude Code Works (opens in a new tab); Elite Context Engineering with Claude Code (opens in a new tab)).

  • Context engineering — read this when your agent is slowing down or wasting tokens: the R&D framework, MCP token waste, priming, bundles, splintering.
  • Prompt engineering — read this when you want better output from any prompt: the format levels, input-workflow-output, XML, examples, anti-overengineering, hallucination control.

Two companion pages add focused guidance:

Resources

Practitioner videos

The videos behind the pages in this section, each cited inline where its claims are used.

Which design principles shape Claude Code?

From How Claude Code Works

Source: How Claude Code Works (opens in a new tab)

  1. Trust in the model. When in doubt, rely on the model.
  2. Simple design wins. One while loop. Flat is better than nested.
  3. Bash is all you need. Minimize tool count. Lean on universal tools.
  4. Context management matters. The longer the context, the stupider the agent.
  5. Different perspectives matter. No single best coding agent. Mix and match.

From the agentic coding framework

  1. The Core Four: Context, Model, Prompt, Tools. Master these fundamentals.
  2. R&D Framework: Reduce and Delegate to manage context.
  3. Focused agent = performant agent. One agent, one purpose.
  4. Consistency beats complexity. Same prompt formats, same section names, every time.
  5. Better > More > Custom. Progress through the agent sophistication ladder.
  6. In-loop > Out-loop > ZTE. Graduate from babysitting to autonomous agents.

From founder workflows

Source: Master Claude Code: Proven Daily Workflows from 3 Founders (opens in a new tab)

  1. Explore, Plan, Execute. Never skip to execute.
  2. Double-escape is your best friend. Fork smart contexts, reuse them.
  3. Fire bad contractors. If Claude goes wrong, don’t correct – restart.
  4. My Developer trick. Separate planner from developer for honest feedback.
  5. Fight enterprise defaults. Explicitly say: no backwards compatibility, no graceful fallbacks, delete dead code.

In this section

  • Context Engineering

    Claude Code techniques for a lean session: the context cost of MCP servers, priming over a growing CLAUDE.md, fresh-agent handoffs, compact versus clear.

  • Prompt Engineering

    For anyone writing prompts for Claude Code: format levels, a section-by-section structure for agentic prompts, and Anthropic's recommended techniques.