Generate Dockerfile

--- name: generate-dockerfile description: Generate production-ready, secure, multi-stage Dockerfile and .dockerignore for any project --- > Vendored from [vfarcic/dot-ai](https://github.com/vfarcic/dot-ai) `shared-prompts/generate-dockerfile.md` (MIT, Copyright (c) 2025 Viktor Farcic). Original author: Viktor Farcic. # Generate Production-Ready Dockerfile Generate an optimized, secure, multi-stage Dockerfile and .dockerignore for the current project by analyzing its structure, language, framework, and dependencies. ## Instructions You are helping a developer containerize their application for production deployment. Your task is to analyze the project structure and generate two files: 1. **Dockerfile**: Production-ready, multi-stage build with security best practices 2. **.dockerignore**: Optimized build context configuration ## Critical Principles These are non-negotiable rules that override all other guidance. ### Verify Everything Before Adding It **ABSOLUTE RULE**: Before adding ANY instruction, configuration, or feature to the Dockerfile, verify it by examining the actual codebase. **Required Process**: 1. **Identify** what you think should be added 2. **Search the codebase** to verify it exists or is actually needed 3. **Only add if verified** - if you can't find evidence in the code, don't add it 4. **When uncertain, ask the user** - if you cannot deduce something from the codebase analysis, ask the user rather than guessing **Never assume. Always verify. Ask when uncertain. Evidence-based Dockerfiles only.** **Thoroughness over speed**: Shallow analysis leads to broken Dockerfiles. Before generating anything: - Read the actual source files, not just file names or directory listings - Search for patterns multiple times with different queries if needed - Trace the application entry point through its imports and dependencies - Don't stop at the first search result - investigate thoroughly - If analysis feels quick, you probably missed something A correct Dockerfile that took longer to generate is far better than a fast but broken one. Spend the time upfront. ### Multi-Architecture Support **REQUIREMENT**: Ensure all Dockerfile instructions support multiple architectures (amd64, arm64, etc.). **Apply to**: - Base image selection: Use multi-arch official images - Binary downloads: Detect architecture dynamically, never hardcode (amd64, x86_64, etc.) - System package installation: Use package manager (automatically handles architecture) - Build commands: Ensure cross-platform compatibility **The Dockerfile must build successfully on different CPU architectures without modification.** ### NEVER Add HEALTHCHECK **ABSOLUTE PROHIBITION**: DO NOT add HEALTHCHECK instruction under ANY circumstances. **Why**: - Health endpoints are application-specific and cannot be verified from codebase analysis - Adding unverified health checks will cause containers to be marked unhealthy incorrectly - Users will add their own HEALTHCHECK if their application has health endpoints **If you add HEALTHCHECK, you are violating the "verify everything" principle.** --- ## Best Practices Reference These are best practices to consider when generating the Dockerfile. **Apply them when relevant to the project** - not every practice applies to every situation: - Package manager flags depend on which package manager is used (apt-get vs apk vs others) - Language-specific guidance applies only to that language - The "verify everything" principle overrides all: if a practice doesn't fit the project, skip it Use this section as guidance during generation and a reference for validation. ### Security | Practice | Description | |----------|-------------| | **Non-root user** | Create and run as a dedicated user (UID 10001+), never run as root | | **Pin image versions** | Use specific tags like `node:20-alpine`, never `:latest` | | **Official images** | Prefer Docker Official Images or Verified Publishers from trusted sources | | **No secrets in image** | Never embed credentials, API keys, or passwords in Dockerfile or ENV instructions | | **No sudo** | Don't use sudo in containers; switch USER explicitly when root access is needed | | **Minimal packages** | Only install packages that are actually required for the application | | **--no-install-recommends** | Use this flag with apt-get to prevent installing optional packages | | **COPY over ADD** | Always use COPY unless you specifically need ADD's tar extraction; never use ADD with URLs | | **No debugging tools** | Avoid installing curl, wget, vim, netcat in production images unless required by the application | | **Clean in same layer** | Remove package manager caches in the same RUN command as installation | | **Executables owned by root** | Application binaries should be owned by root but executed by non-root user | ### Image Selection | Practice | Description | |----------|-------------| | **Minimal base images** | Prefer alpine, slim, distroless, or scratch over full distribution images | | **Multi-stage builds** | Always separate build dependencies from runtime; build stage → runtime stage | | **Match language needs** | Compiled languages → distroless/scratch; Interpreted → slim/alpine with runtime | | **Derive version from project** | Get language version from project files (package.json engines, go.mod, etc.) | ### Build Optimization | Practice | Description | |----------|-------------| | **Layer caching** | Copy dependency manifests (package.json, go.mod) before source code | | **Combine RUN commands** | Chain related commands with `&&` to reduce layers and enable cleanup | | **Explicit COPY** | Never use `COPY . .`; explicitly copy only required files and directories | | **Order by change frequency** | Place stable instructions first (base image, deps) and volatile ones last (source code) | | **Production dependencies only** | Install only production dependencies, not devDependencies | ### Maintainability | Practice | Description | |----------|-------------| | **Sort arguments** | Alphabetize multi-line package lists for easier maintenance and review | | **Use WORKDIR** | Always use WORKDIR to change directories, never `RUN cd` | | **Exec form for CMD** | Use JSON array format: `CMD ["executable", "arg1"]` for proper signal handling | | **Comment non-obvious decisions** | Explain why certain choices were made, not what the command does | | **OCI labels** (optional) | Add metadata labels for image management (org.opencontainers.image.*) | --- ## Process ### Step 0: Check for Existing Dockerfile **Before generating anything, check if the project already has a Dockerfile.** 1. Look for `Dockerfile` in the project root (also check for variants like `Dockerfile.prod`) 2. If found, read and store its contents for Step 2 3. Similarly, check for `.dockerignore` and read it if present This determines whether Step 2 will generate new files or improve existing ones. ### Step 1: Analyze Project Structure **Identify the project characteristics through exploration, not pattern matching.** These are analysis goals, not lookup tables. The examples below are illustrative - apply the same analytical approach to ANY language, framework, or toolchain you encounter. 1. **Language Detection**: Explore the project to identify its programming language(s). - Look for dependency manifest files (e.g., `package.json`, `go.mod`, `requirements.txt`, `Cargo.toml`, `Gemfile`, `composer.json`, `mix.exs`, `build.sbt`, etc.) - Examine source file extensions - Read manifest contents to understand the ecosystem - **Principle**: Every language has some form of dependency declaration - find it and read it 2. **Version Detection**: Find the required language/runtime version. - Search manifest files for version constraints or engine requirements - Look for version files (e.g., `.node-version`, `.python-version`, `.ruby-version`, `.tool-versions`) - Check CI configuration files which often specify versions - **If project specifies a version** → use that exact version - **If no version specified** → search online for the current LTS/stable version of that language/runtime - **Principle**: Use the project's required version if specified, otherwise look up the current recommended version - never guess 3. **Framework Detection**: Identify frameworks from dependencies and project structure. - Read the dependency list in manifest files - Look for framework-specific configuration files - Examine the project structure for framework conventions - **Principle**: Frameworks leave fingerprints - configuration files, directory structures, dependencies 4. **Application Type**: Determine what kind of application this is by examining entry points and configuration. - Web server/API: Look for HTTP server setup, route definitions, port binding - CLI tool: Look for argument parsing, command definitions, bin entries - Worker/background job: Look for queue consumers, scheduled tasks - Static site: Look for build output configuration, no server code - **Principle**: The entry point and its imports reveal the application's purpose 5. **Port Detection**: Search for port configuration in source code and configuration files. - Look for environment variable usage (e.g., `PORT`, `HTTP_PORT`) - Search for hardcoded port numbers in server initialization - Check configuration files for port settings - **Only add EXPOSE if you find concrete evidence** 6. **Build Requirements**: Identify how the project is built. - Read the manifest file for build scripts/commands - Identify the build tool (could be language-standard or third-party) - Determine build outputs (compiled binaries, transpiled code, bundled assets) - **Principle**: Every project that needs building has build instructions - find them 7. **System Dependencies**: Critical step - missing runtime binaries cause silent failures. - Search the codebase for code that executes external commands or binaries - Common patterns: shell execution, subprocess calls, exec functions, system calls - For each binary found, verify it's needed at runtime (not just build time) - Consider what the application actually does - does it need CLI tools, database clients, image processors? - **When uncertain whether something is a runtime dependency, ask the user** 8. **Environment Variable Detection**: Critical step - missing env vars cause runtime failures. - Search the codebase for environment variable access (every language has a way to read env vars) - Look for `.env.example`, `.env.sample`, or similar files that document required variables - Check configuration and startup code for env var usage - Determine which vars are required (no default, app fails without) vs optional (has default) - For required vars, set sensible defaults in the Dockerfile - **Principle**: If the code reads an env var, the container probably needs it configured ### Step 2: Generate or Improve Dockerfile **If no existing Dockerfile** → Generate a new multi-stage Dockerfile using the patterns below. **If existing Dockerfile found** → Analyze it against the best practices and checklists below, then improve: 1. **Evaluate against checklists** - Check each item in the Builder and Runtime checklists 2. **Identify issues** - Security problems (running as root, :latest tags), missing optimizations (no multi-stage, COPY . .), maintainability issues 3. **Preserve intentional customizations** - Comments explaining decisions, custom configurations, environment-specific settings 4. **Edit to fix issues** - Apply best practices while keeping the existing structure where it's already correct 5. **Explain changes** - When presenting the improved Dockerfile, briefly note what was changed and why Use the patterns and checklists below for both generation and validation. **The examples below show structural patterns, not copy-paste templates.** Adapt the pattern to whatever language, package manager, and build tool the project uses. #### Stage 1: Builder ```dockerfile # Build stage - use an image with build tools for this language FROM :- AS builder WORKDIR /app # PATTERN: Copy dependency manifests FIRST for layer caching # Examples: package.json, go.mod, requirements.txt, Gemfile, Cargo.toml, pom.xml COPY ./ # PATTERN: Install dependencies, clean cache in same layer # Use whatever package manager the project uses RUN && \ # PATTERN: Copy only the source files needed for build # Never use "COPY . ." - be explicit about what's needed COPY ./ COPY ./ # PATTERN: Run the project's build command RUN ``` **Builder stage checklist**: - [ ] Named stage (`AS builder`) - [ ] Base image appropriate for the language (with build tools) - [ ] Version derived from project files (not assumed) - [ ] Dependency manifests copied before source code - [ ] Dependencies installed with cache cleanup in same RUN - [ ] Only required files copied (never `COPY . .`) - [ ] Build command matches what the project actually uses #### Stage 2: Runtime ```dockerfile # Runtime stage - use minimal image appropriate for the language # Compiled languages: consider distroless, scratch, or alpine # Interpreted languages: use slim or alpine variant with runtime only FROM : WORKDIR /app # PATTERN: Create non-root user (syntax varies by base image) # Alpine uses addgroup/adduser, Debian uses groupadd/useradd RUN && \ # PATTERN: Copy ONLY runtime artifacts from builder # What you copy depends on the language: # - Compiled: just the binary # - Interpreted: built output + runtime dependencies + minimal config COPY --from=builder ./ COPY --from=builder ./ # PATTERN: Set ownership to non-root user RUN chown -R : /app # Switch to non-root user BEFORE exposing ports or setting CMD USER # Only if port was verified during analysis EXPOSE # PATTERN: Use exec form for proper signal handling # The command depends on how this application runs CMD ["", ""] ``` **Runtime stage checklist**: - [ ] Minimal base image (alpine/slim/distroless/scratch as appropriate) - [ ] Non-root user created (UID 10001+) - [ ] Only runtime artifacts copied from builder - [ ] No source code, tests, build tools, or dev dependencies - [ ] Proper ownership set - [ ] USER directive before CMD - [ ] EXPOSE only if port was verified in analysis - [ ] CMD in exec form (JSON array) #### System Package Installation Pattern When system packages are required, use the package manager appropriate for your base image. The principle is always the same: **install only what's needed and clean the cache in the same layer**. Common examples (adapt to your base image's package manager): ```dockerfile # apt-get (Debian, Ubuntu) RUN apt-get update && \ apt-get install -y --no-install-recommends \ package1 \ package2 && \ rm -rf /var/lib/apt/lists/* # apk (Alpine) RUN apk add --no-cache \ package1 \ package2 # yum/dnf (RHEL, Fedora, CentOS) RUN yum install -y \ package1 \ package2 && \ yum clean all && \ rm -rf /var/cache/yum ``` **Package installation checklist**: - [ ] Used the correct package manager for the base image - [ ] Used flags to skip optional/recommended packages where available - [ ] Packages sorted alphabetically for maintainability - [ ] Cache cleaned in same RUN command - [ ] Only packages actually required by the application ### Step 3: Create or Improve .dockerignore **If no existing .dockerignore** → Generate a minimal one based on the Dockerfile. **If existing .dockerignore found** → Review it against the Dockerfile's COPY commands: 1. Remove redundant exclusions (directories not copied by Dockerfile anyway) 2. Add missing security exclusions (secrets inside copied directories) 3. Keep it minimal (~10-15 lines) **Generate a MINIMAL .dockerignore file based on the Dockerfile.** Since the Dockerfile uses **explicit COPY commands** (not `COPY . .`), .dockerignore serves a limited purpose: 1. **Security** - Exclude secret patterns that could exist INSIDE directories being copied 2. **Performance** - Exclude large directories that slow down build context transfer #### Process 1. Review your Dockerfile's COPY commands - what directories does it copy? 2. Identify security risks inside those directories (secret files that could accidentally exist) 3. Identify large directories in the project (>1MB) that slow context transfer 4. Exclude ONLY those items #### What NOT To Exclude **DO NOT exclude directories that aren't copied by your Dockerfile!** If your Dockerfile doesn't copy a directory, excluding it in .dockerignore is pointless redundancy. #### Target Size **~10-15 lines maximum.** If your .dockerignore exceeds 20 lines, you're likely adding unnecessary exclusions. ### Step 4: Build, Test, and Iterate **Purpose**: Verify the Dockerfile works before presenting to user. A Dockerfile isn't done until it's validated. #### 4.1 Build Build the image to verify the Dockerfile syntax and instructions are correct: ```bash docker build -t [project-name]-validation . ``` - If build succeeds → proceed to run - If build fails → analyze the error, fix Dockerfile, retry #### 4.2 Run Start a container to verify the application runs: ```bash docker run -d --name [project-name]-test [project-name]-validation sleep 5 # Allow startup time ``` Check container state: ```bash docker inspect --format='{{.State.Status}}' [project-name]-test docker inspect --format='{{.State.ExitCode}}' [project-name]-test ``` **Expected behavior depends on application type** (determined in Step 1): - **Services** (web servers, APIs, workers): Container should still be running - **CLI tools / one-shot commands**: Container should have exited with code 0 If container crashed or exited unexpectedly → proceed to log analysis to understand why. #### 4.3 Log Analysis Capture and analyze container logs: ```bash docker logs [project-name]-test 2>&1 ``` **Analyze logs using your knowledge of the project from Step 1.** You know: - What language and framework this is - What the application is supposed to do - What dependencies it requires - What a successful startup looks like for this type of application Use this context to determine if the logs indicate: - The application started correctly, OR - Something is wrong (errors, crashes, missing dependencies, permission issues, etc.) If logs indicate a problem → identify root cause, fix Dockerfile or .dockerignore, retry. #### 4.4 Linting (if available) If `hadolint` is installed, run it to catch Dockerfile best practice issues: ```bash hadolint Dockerfile ``` - If hadolint is not installed → skip this check - If hadolint reports issues → evaluate each issue, fix if appropriate, retry - Some hadolint warnings may be intentional (use judgment based on project context) #### 4.5 Security Scan (if available) If `trivy` is installed, scan the built image for vulnerabilities: ```bash trivy image --severity HIGH,CRITICAL [project-name]-validation ``` - If trivy is not installed → skip this check - If trivy reports HIGH/CRITICAL vulnerabilities in the base image → consider if a different base image version or variant would help - If vulnerabilities are in application dependencies → note them for the user but don't block (dependency updates are outside Dockerfile scope) #### 4.6 Iterate If any validation step fails: 1. **Analyze** the specific error message or behavior 2. **Identify root cause** - common issues include: - Missing file → incorrect COPY command or overly aggressive .dockerignore - Missing dependency → system package not installed - Permission denied → ownership or USER directive issue - Module not found → build step incomplete or wrong files copied - Hadolint warning → Dockerfile best practice issue 3. **Fix** the appropriate file (Dockerfile or .dockerignore) 4. **Retry** from step 4.1 **Maximum 5 iterations.** If still failing after 5 attempts: - Stop and present current state to user - Explain what's failing and what fixes were attempted - Ask for guidance #### 4.7 Cleanup **Always clean up after validation**, whether successful or not: ```bash docker stop [project-name]-test 2>/dev/null || true docker rm [project-name]-test 2>/dev/null || true docker rmi [project-name]-validation 2>/dev/null || true ``` Only proceed to present the Dockerfile to user after: - All validation steps pass, AND - Cleanup is complete --- ## Output Format ### For New Dockerfiles (no existing file) **Present both files to the user:** 1. **Dockerfile** with clear comments explaining each section 2. **.dockerignore** with organized sections **After generating, provide:** - Brief explanation of design choices (base images, build stages, security measures) - Build command: `docker build -t [project-name] .` - Run command: `docker run -p [port]:[port] [project-name]` - Image size expectations ### For Improved Dockerfiles (existing file found) **Present the improved files with a summary of changes:** 1. **Dockerfile** - the improved version 2. **Changes made** - brief list of what was changed and why: - Security fixes (e.g., "Added non-root user - was running as root") - Optimization improvements (e.g., "Added multi-stage build to reduce image size") - Best practice updates (e.g., "Changed CMD to exec form for signal handling") 3. **Preserved** - note any intentional customizations that were kept 4. **.dockerignore** - improved version if changes were needed ### For Both Cases **Recommended next steps** (the Dockerfile has already been validated): - Integrate into CI/CD pipeline - Commit to version control --- ## Success Criteria ### Dockerfile Checklist - [ ] Builds successfully without errors - [ ] Uses multi-stage build (builder → runtime) - [ ] Runs as non-root user (UID 10001+) - [ ] Uses pinned version tags (no `:latest`) - [ ] Uses minimal base images (alpine/slim/distroless) - [ ] Copies dependency manifests before source (layer caching) - [ ] Uses explicit COPY (no `COPY . .`) - [ ] Combines RUN commands with `&&` - [ ] Cleans package manager caches in same layer - [ ] Uses `--no-install-recommends` (if apt-get used) - [ ] Uses exec form for CMD (`["executable", "arg"]`) - [ ] No debugging tools unless required - [ ] No secrets or credentials embedded ### .dockerignore Checklist - [ ] Minimal size (~10-15 lines) - [ ] Excludes secrets inside copied directories - [ ] Excludes large unnecessary directories - [ ] Does NOT exclude directories not copied by Dockerfile ### Validation Checklist (Step 4) - [ ] Image builds successfully - [ ] Container starts without crashing - [ ] Logs show no errors indicating application failure - [ ] Hadolint passes (if installed) - [ ] Trivy shows no critical base image vulnerabilities (if installed) - [ ] Test container and image cleaned up **Do not present Dockerfile to user until all validation checks pass.** --- ## Example Workflows ### New Dockerfile (no existing file) 1. **Check**: "No existing Dockerfile found. Will generate new one." 2. **Explore**: "Let me find the dependency manifest... found ``. Reading it to understand the ecosystem." 3. **Identify**: "This is a `` project using ``. The manifest indicates version ``." 4. **Trace**: "The entry point is ``. Following imports to understand runtime needs." 5. **Structure**: "Multi-stage build: builder stage needs ``, runtime stage needs only ``." 6. **Dependencies**: "Searching for external binary usage... found ``. This needs to be in the runtime image." 7. **Generate**: "Create Dockerfile and .dockerignore. Check against best practices checklists." 8. **Build & Test**: "Building image... Running container... Checking logs..." 9. **Iterate** (if needed): "Build failed due to missing package. Adding to Dockerfile and retrying..." 10. **Cleanup & Present**: "Validation passed. Removing test artifacts. Here's your Dockerfile." ### Improving Existing Dockerfile 1. **Check**: "Found existing Dockerfile. Reading it to analyze..." 2. **Analyze project**: Same exploration as above to understand what the Dockerfile should do. 3. **Evaluate**: "Checking existing Dockerfile against best practices checklists..." - "❌ Running as root - no USER directive" - "❌ Using :latest tag instead of pinned version" - "✅ Multi-stage build already in place" - "✅ Dependency manifests copied first" 4. **Preserve**: "Keeping custom ENV variables and the specific port configuration - these appear intentional." 5. **Improve**: "Adding non-root user, pinning image version to match project requirements." 6. **Build & Test**: "Building improved image... Running container... Checking logs..." 7. **Iterate** (if needed): "Container crashed - logs show permission error. Fixing ownership and retrying..." 8. **Cleanup & Present**: "Validation passed. Removing test artifacts. Here are the improvements." **Key mindset**: Investigate the actual project rather than matching against templates. Every project is unique. Don't present until validated.