cn-check（中国检查）

cn check — Local AI Agent Checks

Run AI-powered code checks locally against your working tree changes using the Continue CLI. Each check is an agent (defined in markdown) that reviews your diff, identifies issues, and optionally suggests fixes as a patch.

When to Use

• User asks to run AI checks on their code changes
• User wants to set up cn check in a project
• User needs to create custom check agents
• User wants to apply AI-suggested fixes locally
• User asks about Continue CI or agent-based code review

Installation

Prerequisites

• Node.js 18+
• A git repository with uncommitted or branched changes

Install the CLI

npm install -g @continuedev/cli

Authenticate (required for Hub checks, optional for local-only)

cn login

This opens a browser for authentication. After login, Hub-configured checks are available automatically.

Usage

Basic: Run all discovered checks

cn check

This auto-detects checks from three sources (in priority order):

1. Hub API — checks configured for your repo on continue.dev
2. Local agents — markdown files in .continue/agents/*.md

Specify agents explicitly

# Run a single local agent
cn check --agent .continue/agents/security-review.md

# Run a Hub-published agent
cn check --agent myorg/code-style

# Run multiple agents
cn check --agent .continue/agents/security.md --agent .continue/agents/docs.md

Compare against a specific base branch

cn check --base develop

Default: auto-detects main or master.

Output formats

# JSON output (for CI pipelines or scripting)
cn check --format json

# Unified patch output (pipe to git apply)
cn check --patch | git apply

# Stop on first failure
cn check --fail-fast

Auto-fix mode

cn check --fix

Runs all checks, then applies any suggested patches directly to the working tree. Patches that conflict are reported but skipped.

Creating a Check Agent

Create a markdown file at .continue/agents/<name>.md:

# Security Review

You are a security reviewer. Examine the code changes for:

- SQL injection vulnerabilities
- XSS risks in user-facing output
- Hardcoded secrets or credentials
- Insecure use of eval() or similar

If you find issues, edit the files to fix them. If everything looks good, say so and exit.

The agent receives:

• The full diff against the base branch
• A list of changed files
• Access to read/edit files in a temporary worktree

Any edits the agent makes are captured as a patch and reported as a "fail" with suggested changes.

How It Works

1. Diff — Computes git diff <base>...HEAD to find changed files
2. Resolve — Discovers which checks to run (Hub, local, or --agent flags)
3. Worktree — Creates a temporary git worktree per check for isolation
4. Run — Forks a worker process per check; the agent runs with full tool access
5. Capture — After the agent finishes, captures git diff in the worktree as a patch
6. Report — Renders results: pass (no changes), fail (patch produced), or error

Checks run in parallel by default. Use --fail-fast for sequential execution that stops on first failure.

Output

Interactive terminal (TTY)

A live-updating table shows check progress:

cn check  -  3 checks against main  -  5 changed files

Check               Status         Time
--------------------------------------------
Security Review     * Running       12s
Code Style          Pass            8s
Documentation       Pending         -

When complete, a full report prints with pass/fail status, agent output, and suggested patches.

JSON output (--format json)

{
  "checks": [
    {
      "agent": ".continue/agents/security.md",
      "name": "security",
      "status": "pass",
      "patch": "",
      "output": "No security issues found.",
      "duration": 8.2
    }
  ],
  "summary": {
    "total": 1,
    "passed": 1,
    "failed": 0,
    "errored": 0
  }
}

CLI Reference

cn check [options]

Options:
  --base <branch>     Base branch for diff (default: auto-detect)
  --format <format>   Output format: text or json (default: text)
  --fix               Apply suggested fixes to working tree
  --patch             Output unified patch (pipe to git apply)
  --fail-fast         Stop after first failing check
  --agent <agent>     Agent to run (hub slug or local path, repeatable)
  --config <path>     Path to config file
  --org <slug>        Organization slug
  --verbose           Enable debug logging

Troubleshooting