[TERMINAL · SKILLS]
> mounting /skills...
> indexing skill manifests...
> linking agents: claude · codex · gemini · cursor
> ready.
[░░░░░░░░░░░░░░░░░░░░░░░░░░░░] 0%
Skills/Cline
>

Cline

Cline is an open-source autonomous coding agent that runs in VS Code, JetBrains IDEs, the terminal (npm package `cline`) and a desktop app, and edits files, runs commands and uses MCP tools with your approval. Use when a user asks to set up Cline, run it headless in CI, choose a model provider, write .clinerules, connect an MCP server, switch between Plan and Act mode, control auto-approve or YOLO mode, or hand a multi-file refactor to an agent.

#ai-coding#autonomous-agent#cline#mcp#cli
terminal-skillsv1.1.0
Works with:claude-codeopenai-codexgemini-clicursor
Source
Trust Score
93/ 100
4.53×
Impact

Validation

Score is stale (SKILL.md changed since last review)
Quality
93/ 100
Does it follow best practices?
5 PASS · 1 WEAK
Security
Passed
No known issues
Content review + injection scan
Impact
4.53×
17% → 77% agent success
Avg across 3 eval scenarios
Scored 10/24/2026 · skill v1.1.0
$
✓ Installed cline v1.1.0

Getting Started

  1. Install the skill using the command above
  2. Open your AI coding agent (Claude Code, Codex, Gemini CLI, or Cursor)
  3. Reference the skill in your prompt
  4. The AI will use the skill's capabilities automatically

Example Prompts

  • "Review the open pull requests and summarize what needs attention"
  • "Generate a changelog from the last 20 commits on the main branch"

Documentation

Overview

Cline is an open-source coding agent that reads a project, edits files across it, runs shell commands and watches their output, and can call MCP tools. It ships as a VS Code extension (saoudrizwan.claude-dev), a JetBrains plugin, a CLI (npm install -g cline), a desktop app for macOS and Windows, and an SDK (@cline/sdk). All surfaces share one agent core, so rules, skills, MCP servers, checkpoints and provider settings behave the same everywhere. As of October 2026 the CLI is at 3.0.x.

Two modes matter: Plan mode explores and proposes a strategy without changing anything; Act mode executes it. In the IDE every edit and command asks for approval unless you enable auto-approve. In the CLI the default is different (see Guidelines).

Instructions

Install

bash
# CLI
npm install -g cline
cline version

# VS Code extension
code --install-extension saoudrizwan.claude-dev

The JetBrains plugin comes from the JetBrains Marketplace (search "Cline"). The desktop app is downloaded from cline.bot/desktop.

Choose a provider and model

Cline works with its own hosted provider (sign in with cline auth cline), a ChatGPT subscription (openai-codex), or your own key for Anthropic, OpenAI, Google Gemini, OpenRouter, AWS Bedrock, GCP Vertex, Groq, Cerebras, local Ollama or LM Studio, and any OpenAI-compatible endpoint.

bash
cline auth                                   # interactive setup
cline auth --provider anthropic --apikey "$ANTHROPIC_API_KEY" --modelid claude-sonnet-4-6
cline -P openrouter -m google/gemini-3.1-pro-preview -k "$OPENROUTER_API_KEY" "Set up Storybook"

In the IDE extensions, pick the provider and model in the Cline panel settings instead. Provider keys are stored in ~/.cline/data/settings/providers.json.

Run the CLI

bash
cline                                        # interactive session (TUI)
cline -p "Plan a migration from Express to Hono"      # plan mode, no edits
cline --auto-approve false "Fix the failing tests"    # ask before every tool call
git diff origin/main | cline "Review these changes for bugs"
cline --json "List all TODO comments"        # NDJSON events for scripts
cline --yolo --timeout 900 "Run tests and fix failures"   # unattended, CI only

cline history lists past sessions and cline --id 8f3a91c2 resumes one. cline doctor diagnoses configuration problems.

Give it project rules

Rules are markdown files Cline loads on every task. Put them in .clinerules/ (or .cline/rules/) at the repository root; commit them. Cline also reads AGENTS.md, .cursorrules and .windsurfrules. Personal rules go in ~/Documents/Cline/Rules/ or ~/.cline/rules/; workspace rules win on conflict. Type /newrule in a session to have Cline write one for you.

markdown
# .clinerules/testing.md
- Run `npm test -- --run` after every change under src/ and fix failures before finishing.
- Never edit files in legacy/ (scheduled for removal in Q1).
- Use Vitest, not Jest, for new tests.

Add MCP servers

bash
cline mcp                                    # list, add, edit, disable, authorize, delete
cline mcp add files --yes -- npx -y @modelcontextprotocol/server-filesystem ./docs
cline mcp add docs https://mcp.context7.com/mcp --transport http --yes
cline mcp remove files

The CLI config is ~/.cline/data/settings/cline_mcp_settings.json (override with CLINE_MCP_SETTINGS_PATH). In an IDE, open the MCP Servers icon, then Configure. An entry looks like this; set "type": "streamableHttp" for remote servers because an omitted type means legacy SSE:

json
{
  "mcpServers": {
    "postgres-readonly": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres", "postgresql://readonly@localhost:5432/shop"],
      "disabled": false,
      "autoApprove": []
    }
  }
}

Useful slash commands

/newtask starts a fresh task with distilled context, /smol compresses history, /newrule creates a rule, /deep-planning investigates then plans before coding, /undo (CLI) or checkpoints (IDE) roll back the agent's work.

Examples

Example 1: Review a pull request in CI

User: "Have Cline comment on every PR with a review of the diff."

yaml
# .github/workflows/cline-review.yml
name: Cline review
on: pull_request
permissions: { contents: read, pull-requests: write }
jobs:
  review:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with: { fetch-depth: 0 }
      - run: npm install -g cline
      - run: cline auth --provider anthropic --apikey "$ANTHROPIC_API_KEY" --modelid claude-sonnet-4-6
        env: { ANTHROPIC_API_KEY: "${{ secrets.ANTHROPIC_API_KEY }}" }
      - run: git diff origin/${{ github.base_ref }}...HEAD | cline -p "Review for bugs and missing tests" > review.md
      - run: gh pr comment ${{ github.event.number }} --body-file review.md
        env: { GH_TOKEN: "${{ secrets.GITHUB_TOKEN }}" }

Plan mode (-p) keeps the reviewer from editing the checkout. The result is a markdown review posted as a PR comment.

Example 2: Refactor with a plan first

User: "Migrate our API routes from Express to Hono, keep the route structure."

In the IDE, stay in Plan mode and ask Cline to read src/routes/ and list the files it would touch. Review the plan, then switch to Act mode. Cline edits each route, updates imports, and runs npm test, showing every diff for approval. If a step goes wrong, restore the checkpoint taken before it and rephrase the instruction instead of arguing in the same thread.

Guidelines

  • The CLI defaults to Act mode with auto-approve on. Use --auto-approve false or -p when you want a human in the loop, and use --yolo only in a throwaway container or CI runner: it is the flag for unsupervised runs and is dangerous outside a sandbox (in the IDE, YOLO mode approves file changes and commands anywhere on the machine).
  • In the IDE, start with auto-approve limited to reading project files. Enable edits only when checkpoints are available.
  • .clineignore is being removed and was never a security boundary; do not rely on it to hide secrets. Keep secrets out of the workspace and out of rules files.
  • Pass API keys through environment variables or CI secrets, never in committed files; -k also saves the key as the provider's key.
  • Only add MCP servers you trust, and keep autoApprove limited to read-only tools.
  • Break big tasks into steps; long sessions lose detail. Use /newtask or /smol before the context fills.
  • Focus Chain todo lists and "Explain Changes" are deprecated; ask Cline in plain words to explain its changes.