AI & LLMs · Guide · AI & Prompt Tools
How to Use Cline
Installing the Cline extension, task-based workflows, tool permissions, browser use, terminal commands, approval modes.
Cline (formerly Claude Dev) is an autonomous coding agent that lives in a VS Code side panel. It reads your files, edits them, runs terminal commands, and can drive a browser — all with a human-in-the-loop approval gate so you see every diff before it lands.
Advertisement
What Cline actually is
Cline is a VS Code extension that wraps a frontier LLM in a tool-using agent loop. Each turn it picks one tool — read file, write file, run command, search, fetch URL, use a browser — and you approve or reject the action. The model you bring supplies the intelligence; Cline supplies the plumbing, a diff UI, and the permission prompts.
Installing
Install the Cline extension from the VS Code Marketplace (the publisher is saoudrizwan). Open the side panel, pick a provider — Anthropic, OpenRouter, AWS Bedrock, GCP Vertex, or a local Ollama endpoint — and paste a key. OpenRouter is the easy button if you want to try Claude, GPT, and Gemini without three separate accounts.
A first task
Open the Cline panel, type something concrete: “add a /api/users/:id route with zod input validation, a happy-path test, and update the OpenAPI file.” Cline will plan, list files, open the ones it needs, and propose edits in a diff view. Hit approve for each step, or toggle Auto-approve for read-only actions so it does not pause on every cat.
Plan vs Act mode
Plan mode is chat-only — the agent can read but not write. Use it to negotiate the design first (“where should this live, which existing helper do we reuse”) and then switch to Act to execute. Skipping Plan on anything non-trivial is the most common way Cline users burn tokens on a first draft they have to throw away.
Configuration that matters
Set Auto-approve for reads and shell commands you trust (lint, test, typecheck), but keep rm, git push, and writes outside the workspace gated. Add a .clinerules file — it is injected into the system prompt, and it is how you tell Cline “always use pnpm, never npm” or “run our make check before claiming done.” Cap context window spend via the model selector if you are using Sonnet on a long session.
When Cline shines
Medium-to-large features where you want an agent that can run the test suite, read the failure, and keep iterating — all while you review diffs visually in VS Code. The browser tool is genuinely useful for checking a running frontend without tab-switching.
When not to use it
If you live in JetBrains or Neovim, Cline is not for you — try Continue or Aider instead. It is also overkill for single-line tweaks where Copilot-style inline completion is faster. And watch the bill: running Claude Sonnet on a repo with a big context can cost more in one afternoon than a month of Copilot, so set a provider spend limit on day one.
Use these while you read
Tools that pair with this guide
- System Prompt BuilderCompose a focused system prompt from a role, tone, constraints, and output format — copy-ready for any LLM.AI & Prompt Tools
- AI Prompt GeneratorTurn a vague idea into a structured prompt. Pick role, task, context, constraints, and output format. Works with ChatGPT, Claude, and Gemini.AI & Prompt Tools
- AI Token CounterEstimate tokens, characters, words, and approximate API cost for GPT-4o, GPT-4, Claude, and Gemini — before you hit send.AI & Prompt Tools
- AI Prompt LibraryBrowse a curated catalog of prompt templates for writing, coding, marketing, and research. One click to copy.AI & Prompt Tools
Advertisement
Continue reading
- AI & LLMsGitHub Copilot Pricing and ComparisonCompare free vs paid GitHub Copilot tiers and analyze it against ChatGPT, Cursor, and Tabnine. Find the best value plan instantly with this free online guide.
- AI & LLMsGitHub Copilot Features and CapabilitiesTest what Copilot really does — code accuracy, scope limits, debugging, web dev, legacy code, tests, docs, team customization. Free guide, no sign-up.
- AI & LLMsGitHub Copilot Security and Data HandlingAudit where your code goes, who sees it, training-data policy, network needs, and what happens when Copilot suggests broken code. Free, no sign-up.
- AI & LLMsAI Fluency SkillsThe 8 sub-skills of AI fluency: prompt structure, model selection, tool use, quality calibration, iteration, context management, cost awareness, privacy.
- AI & LLMsAnthropic Skills ExplainedSkills as Anthropic's answer to Custom GPTs — markdown-defined, version-controlled in git, work in terminal. Anatomy + Skills vs Custom GPTs.
- AI & LLMsKimi K2 vs DeepSeek V3Two open-weight Chinese flagships. Kimi K2 = 1M context, DeepSeek V3.2 = top-tier reasoning + coding. Pick by use case.