Type: url
Original: https://www.anthropic.com/engineering/claude-code-best-practices
Status: generated
Summary
The Claude Code documentation’s best-practices page, organising its advice around one constraint: the context window fills quickly and the model’s performance degrades as it does. It covers giving the agent a check it can run so the loop closes without a human, separating exploration and planning from implementation via plan mode, writing a short CLAUDE.md and pruning it, configuring permissions and sandboxing, and extending the tool with hooks, skills, subagents and plugins — then turns to managing a session (course-correcting, clearing context, rewinding to checkpoints) and to scaling out through non-interactive mode, parallel sessions, fan-out and an adversarial review step.