# Automation & CI/CD

Claude Code supports a headless mode you can use in scripts, CI/CD pipelines and automated workflows with no human interaction. This article covers the complete flag reference and several real scenarios.

---

## 1. Basic Usage

### One-shot execution (-p mode)

```bash
# Basic run
claude -p "Analyse this project's architecture"

# JSON output
claude -p "List every TODO comment" --output-format json

# Cap the number of turns
claude -p "Write unit tests for UserService" --max-turns 5

# Pick a model (claude-sonnet-5 for everyday work; 4.x is still on sale)
claude -p "Review the code for security issues" --model claude-opus-5
```

### Restricting the available tools

```bash
# Read-only analysis (no writes, no command execution)
claude -p "Analyse code quality" --allowedTools Read,Glob,Grep

# Read and write (no command execution)
claude -p "Refactor this file" --allowedTools Read,Write,Edit,Glob,Grep

# All tools (controlled environment)
claude -p "Fix the lint errors" --allowedTools Read,Write,Edit,Bash,Glob,Grep
```

### Skipping permission prompts

```bash
# Only in a safe, isolated environment!
claude -p "Fix every lint error and commit" --dangerously-skip-permissions
```

> **Security warning**: `--dangerously-skip-permissions` skips every permission confirmation. **Use it only inside a Docker container or an isolated CI environment.**

---

## 2. Complete Flag Reference

### Execution control

| Flag | What it does | Example |
|------|------|------|
| `-p "prompt"` | Headless mode, run a single task | `claude -p "analyse the architecture"` |
| `--bare` | Minimal mode, skips hooks/LSP/plugins | `claude --bare -p "..."` |
| `--max-turns N` | Cap the number of interaction turns | `--max-turns 5` |
| `--model MODEL` | Choose a model | `--model claude-opus-5` |
| `--dangerously-skip-permissions` | Skip every permission confirmation | Isolated environments only |

### Output formats

| Flag | What it does | When to use |
|------|------|---------|
| `--output-format text` | Plain text (default) | For humans |
| `--output-format json` | Structured JSON | For scripts |
| `--output-format stream-json` | Streaming JSON | Real-time processing |

### Tool restrictions

| Flag | What it does |
|------|------|
| `--allowedTools Tool1,Tool2` | Allow only the listed tools |

**Available tool names**: `Read`, `Write`, `Edit`, `Bash`, `Glob`, `Grep`, `WebFetch`, `WebSearch`, `Agent`, `NotebookEdit`

### Session management

| Flag | What it does |
|------|------|
| `--session-id ID` | Set the session id |
| `--resume` | Resume a previous session |

### Environment variables

| Variable | What it does |
|------|------|
| `ANTHROPIC_BASE_URL` | API endpoint (QCode.cc: `https://api.qcode.cc/api`) |
| `ANTHROPIC_AUTH_TOKEN` | API key (starts with `cr_`) |
| `CLAUDE_CODE_MAX_TURNS` | Default maximum turns |
| `CLAUDE_CODE_OUTPUT_FORMAT` | Default output format |
| `CLAUDE_MODEL` | Default model |

---

## 3. CI/CD Recipes

### Recipe 1: GitHub Actions — AI code review

```yaml
name: AI Code Review
on: [pull_request]

jobs:
  review:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0  # full history, needed for the diff

      - name: Setup Node.js
        uses: actions/setup-node@v4
        with:
          node-version: '22'

      - name: Install Claude Code
        run: npm install -g @anthropic-ai/claude-code

      - name: AI Code Review
        env:
          ANTHROPIC_BASE_URL: "https://api.qcode.cc/api"
          ANTHROPIC_AUTH_TOKEN: ${{ secrets.QCODE_API_KEY }}
        run: |
          # Collect the files changed in this PR
          FILES=$(git diff --name-only origin/${{ github.base_ref }}...HEAD)

          claude -p "Review the code changes in the following files, focusing on:
          1. Security vulnerabilities (SQL injection, XSS, leaked secrets)
          2. Performance problems (N+1 queries, memory leaks)
          3. Logic errors
          4. Code style issues

          Changed files:
          $FILES" \
            --output-format json \
            --max-turns 3 \
            --model claude-sonnet-5 \
            --allowedTools Read,Glob,Grep \
            > review.json

          echo "Review completed"
          cat review.json | jq -r '.result' || cat review.json
```

### Recipe 2: generate tests automatically

```yaml
name: Auto Generate Tests
on:
  push:
    paths: ['src/**/*.ts', '!src/**/*.test.ts']

jobs:
  generate-tests:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      - uses: actions/setup-node@v4
        with:
          node-version: '22'

      - name: Install dependencies
        run: |
          npm ci
          npm install -g @anthropic-ai/claude-code

      - name: Generate missing tests
        env:
          ANTHROPIC_BASE_URL: "https://api.qcode.cc/api"
          ANTHROPIC_AUTH_TOKEN: ${{ secrets.QCODE_API_KEY }}
        run: |
          claude -p "Look at the .ts files under src/ and write unit tests for the ones that have none.
          Use Vitest + Testing Library.
          Name test files xxx.test.ts and put them next to the source file.
          Target 80%+ coverage." \
            --max-turns 10 \
            --allowedTools Read,Write,Glob,Grep,Bash \
            --dangerously-skip-permissions

      - name: Run tests
        run: npx vitest --run

      - name: Create PR with tests
        if: success()
        run: |
          git config user.name "claude-bot"
          git config user.email "bot@qcode.cc"
          git checkout -b auto-tests-$(date +%s)
          git add '*.test.ts'
          git commit -m "test: auto-generated unit tests" || exit 0
          git push origin HEAD
```

### Recipe 3: a code quality gate

```yaml
name: Code Quality Gate
on: [pull_request]

jobs:
  quality:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      - uses: actions/setup-node@v4
        with:
          node-version: '22'

      - run: npm install -g @anthropic-ai/claude-code

      - name: Quality Analysis
        env:
          ANTHROPIC_BASE_URL: "https://api.qcode.cc/api"
          ANTHROPIC_AUTH_TOKEN: ${{ secrets.QCODE_API_KEY }}
        run: |
          claude -p "Analyse code quality and output JSON:
          {
            \"score\": 0-100,
            \"issues\": [{\"severity\": \"high|medium|low\", \"file\": \"...\", \"description\": \"...\"}],
            \"summary\": \"one-line summary\"
          }

          Scoring:
          - Type safety (20 points)
          - Error handling (20 points)
          - Test coverage (20 points)
          - Readability (20 points)
          - Security (20 points)" \
            --output-format json \
            --max-turns 3 \
            --model claude-sonnet-5 \
            --allowedTools Read,Glob,Grep \
            > quality.json

      - name: Check score
        run: |
          SCORE=$(cat quality.json | jq -r '.result' | jq -r '.score // 0')
          echo "Quality score: $SCORE"
          if [ "$SCORE" -lt 60 ]; then
            echo "Quality gate failed: score $SCORE < 60"
            exit 1
          fi
```

### Recipe 4: generate a changelog

```bash
#!/bin/bash
# generate-changelog.sh — build a changelog from git commits

export ANTHROPIC_BASE_URL="https://api.qcode.cc/api"
export ANTHROPIC_AUTH_TOKEN="cr_your_api_key"

# Commits since the last tag
LAST_TAG=$(git describe --tags --abbrev=0 2>/dev/null || echo "")
if [ -z "$LAST_TAG" ]; then
  COMMITS=$(git log --oneline -20)
else
  COMMITS=$(git log --oneline ${LAST_TAG}..HEAD)
fi

claude -p "Produce a structured changelog from these git commits:

$COMMITS

Format:
## [version] - $(date +%Y-%m-%d)
### Added
### Fixed
### Changed
### Breaking changes (if any)

Write in English, concise and professional." \
  --output-format text \
  --max-turns 2 \
  --model claude-sonnet-5 \
  --allowedTools Read,Glob,Grep
```

### Recipe 5: keep the docs up to date

```bash
#!/bin/bash
# update-docs.sh — refresh the API docs after code changes

export ANTHROPIC_BASE_URL="https://api.qcode.cc/api"
export ANTHROPIC_AUTH_TOKEN="cr_your_api_key"

claude -p "Look at the route files under src/api/ and compare them with the documentation in docs/api.md.
Find API descriptions that are missing or out of date and update docs/api.md.
Keep the existing document format and style." \
  --max-turns 8 \
  --allowedTools Read,Write,Edit,Glob,Grep
```

---

## 4. Docker Isolation

When you use `--dangerously-skip-permissions` in CI/CD, we strongly recommend running inside a Docker container:

```dockerfile
FROM node:22-slim

# Install Claude Code
RUN npm install -g @anthropic-ai/claude-code

# Install project dependencies
WORKDIR /app
COPY package*.json ./
RUN npm ci
COPY . .

# QCode.cc environment
ENV ANTHROPIC_BASE_URL=https://api.qcode.cc/api
# ANTHROPIC_AUTH_TOKEN is injected at runtime

# Run as a non-root user
RUN useradd -m claude
USER claude

CMD ["claude", "-p", "Run a code review", "--dangerously-skip-permissions", "--max-turns", "5"]
```

Run it:

```bash
docker build -t claude-ci .
docker run --rm -e ANTHROPIC_AUTH_TOKEN=cr_your_key claude-ci
```

---

## 5. Session Resumption and Multi-step Pipelines

### Multi-step tasks

```bash
# Step 1: analyse
claude -p "Analyse the project architecture" \
  --session-id "pipeline-42" \
  --output-format json \
  --max-turns 3 \
  --allowedTools Read,Glob,Grep

# Step 2: build a plan from that analysis
claude -p "Based on the previous analysis, draft a refactoring plan" \
  --resume --session-id "pipeline-42" \
  --max-turns 3

# Step 3: execute the plan
claude -p "Carry out the first step of the refactoring plan" \
  --resume --session-id "pipeline-42" \
  --max-turns 10 \
  --allowedTools Read,Write,Edit,Bash,Glob,Grep
```

---

## 6. Controlling Cost

### Strategy 1: cap the turns

```bash
# Three turns is enough for a simple task
claude -p "Quick analysis" --max-turns 3

# Ten at most for a complex one
claude -p "Full refactor" --max-turns 10
```

### Strategy 2: pick the right model

| Scenario | Recommended | Why |
|------|---------|------|
| Code review | Sonnet | Good enough and cheap |
| Security scanning | Opus | Needs deep analysis |
| Bulk formatting | Haiku | Cheapest |
| Test generation | Sonnet | Best value |

```bash
claude -p "Format the code" --model claude-haiku-4-5 --max-turns 3
```

### Strategy 3: restrict tools to cut token usage

```bash
# Read-only analysis → no repeated write/test cycles to pay for
claude -p "Analyse the code" --allowedTools Read,Glob,Grep --max-turns 3
```

---

## 7. Compared with Codex Headless

| | Claude Code -p | Codex headless |
|------|---------------|----------------|
| Flag | `-p "prompt"` | `codex -q "prompt"` |
| Sandbox | Needs Docker isolation | Built-in kernel-level sandbox |
| Output formats | text/json/stream-json | text/json |
| Session resumption | `--resume --session-id` | Not supported |
| Tool restriction | `--allowedTools` | `--approval-mode` |
| Parallel execution | Not supported | `codex cloud exec` |

**Use them together**: Claude Code for analysis and planning (read-only), Codex for execution (fully automatic).

> A QCode.cc plan shares its quota, so switching between the two tools in CI costs nothing.

---

## 8. Best Practice Checklist

1. **Always cap the turns**: use `--max-turns` in CI to prevent runaway execution
2. **Restrict the tools**: expose only what the task needs via `--allowedTools`
3. **Isolate with Docker**: `--dangerously-skip-permissions` must run in a container
4. **JSON output**: prefer `--output-format json` in automation so results are easy to parse
5. **Control cost**: Sonnet for analysis, Haiku for simple tasks
6. **Idempotency**: make sure repeated runs have no side effects
7. **Timeouts**: set a job timeout in the pipeline (10 minutes, say)
8. **Secret management**: keep the API key in CI secrets, never hard-code it

---

## Next Steps

- [Hooks System](/docs/advanced/hooks) — trigger actions automatically
- [CLI Tips](/docs/usage/cli-tips) — more command-line usage
- [Cost Optimization](/docs/usage/cost-optimization) — techniques for controlling spend
- [Complete Claude Code Tutorial](/docs/getting-started/claude-code-tutorial) — from zero to fluent