CI/CD Pipeline Generation from Codebase Analysis: A Complete Guide to Claude Skills

The CI/CD Pipeline Builder skill in alirezarezvani/claude-skills automatically detects your repository's technology stack and generates production-ready GitHub Actions or GitLab CI workflows using zero-dependency Python scripts.

This guide explains how to leverage automated CI/CD pipeline generation from codebase analysis to eliminate manual YAML configuration. The skill resides in the engineering/ci-cd-pipeline-builder directory of the Claude Skills repository and provides a portable, CLI-first solution for transforming repository signals into continuous integration definitions.

How the CI/CD Pipeline Builder Works

The system operates through two core Python modules that form a detection-to-generation pipeline. Both scripts use only Python standard libraries, ensuring compatibility across any environment without package installation overhead.

Stack Detection with stack_detector.py

The stack_detector.py module located at engineering/ci-cd-pipeline-builder/scripts/stack_detector.py performs recursive repository analysis to identify technology signals. It scans for manifest files including package.json, pyproject.toml, go.mod, lock files, and Dockerfile instances to construct a StackReport dataclass.

This report encapsulates detected programming languages, package managers, appropriate CI targets, and common development commands such as lint, test, and build operations. The detector outputs either human-readable text or structured JSON via the --format json flag, enabling programmatic consumption by downstream tools.

Pipeline Generation with pipeline_generator.py

The pipeline_generator.py script at engineering/ci-cd-pipeline-builder/scripts/pipeline_generator.py consumes StackReport data to emit platform-specific YAML configurations. It intelligently selects correct node installers—distinguishing between npm ci, pnpm install, and yarn install—and injects language-specific steps for Node.js, Python, and Go environments.

The generator supports both GitHub Actions and GitLab CI platforms, creating appropriately named jobs such as node-ci, python-ci, and go-ci for GitHub, or node_lint, node_test, and node_build for GitLab configurations. Upon completion, it outputs the YAML to a specified file path and prints a concise PipelineSummary to stdout or stderr based on the selected format mode.

Data Flow Architecture

The detection and generation process follows a strict four-stage pipeline:

1. Detection – stack_detector.py walks the target directory, sets boolean signal flags, and aggregates language-specific commands.
2. Payload – The resulting StackReport serializes to JSON or streams directly to the generator.
3. Generation – pipeline_generator.py reads the payload (or auto-detects via internal logic) and builds YAML strings with platform-specific job definitions.
4. Output – The final workflow writes to .github/workflows/ci.yml or any custom path supplied via --output, accompanied by a generation summary.

Reference templates stored in references/github-actions-templates.md and references/gitlab-ci-templates.md provide the foundational snippets used during generation, while engineering/ci-cd-pipeline-builder/SKILL.md exposes the capability to Claude Code agents through standardized metadata definitions.

Using the CI/CD Pipeline Builder

The following workflows demonstrate practical implementations of CI/CD pipeline generation from codebase analysis for different scenarios and platforms.

Detecting Your Technology Stack

Generate a comprehensive stack report to understand what the detector identifies in your repository:

python3 engineering/ci-cd-pipeline-builder/scripts/stack_detector.py \
  --repo . \
  --format json > stack.json

This command analyzes the current directory, identifies all technology markers, and writes a structured StackReport JSON payload to stack.json for inspection or manual editing before pipeline generation.

Generating GitHub Actions Workflows

Transform an existing stack report into a GitHub Actions workflow file:

python3 engineering/ci-cd-pipeline-builder/scripts/pipeline_generator.py \
  --input stack.json \
  --platform github \
  --output .github/workflows/ci.yml \
  --format text

This creates .github/workflows/ci.yml containing jobs for each detected language and prints a "Pipeline generated" summary to confirm successful creation.

One-Shot Generation for GitLab CI

Execute complete CI/CD pipeline generation from codebase analysis without intermediate files by allowing the generator to handle detection internally:

python3 engineering/ci-cd-pipeline-builder/scripts/pipeline_generator.py \
  --repo . \
  --platform gitlab \
  --output .gitlab-ci.yml

When no JSON input is supplied via --input, the script falls back to its own lightweight detection logic and immediately writes a GitLab CI configuration file.

Integration with Claude Code

Install and invoke the skill directly within Claude Code environments using the standardized skill interface:

/plugin install ci-cd-pipeline-builder@claude-code-skills
/skill ci-cd-pipeline-builder generate --repo /path/to/project --platform github

The SKILL.md metadata file exposes a generate command that orchestrates both detection and generation scripts behind a unified interface, making the tool discoverable through the Claude Code marketplace.

Key Files and Implementation Details

Understanding the repository structure ensures effective customization and troubleshooting:

• engineering/ci-cd-pipeline-builder/scripts/stack_detector.py – Implements repository walking and signal detection logic, producing the StackReport dataclass.
• engineering/ci-cd-pipeline-builder/scripts/pipeline_generator.py – Contains YAML templating engines for both GitHub Actions and GitLab CI platforms.
• engineering/ci-cd-pipeline-builder/README.md – Provides installation instructions for Claude Code, Codex, and OpenClaw integrations.
• engineering/ci-cd-pipeline-builder/SKILL.md – Defines input/output schemas and command mappings for Claude agent consumption.
• engineering/ci-cd-pipeline-builder/references/ – Houses template markdown files for GitHub Actions and GitLab CI syntax reference.

The implementation maintains a zero-dependency architecture using only Python standard libraries, ensuring the tool functions in minimal container environments or fresh virtual machines without pip installation requirements.

Summary

• Automated Detection: The stack_detector.py script identifies languages, package managers, and build tools by scanning for manifest files like package.json and go.mod.
• Multi-Platform Support: pipeline_generator.py creates valid YAML for both GitHub Actions and GitLab CI, selecting appropriate installers and job structures automatically.
• Flexible Workflows: Support for JSON intermediates or one-shot generation accommodates both manual review requirements and fully automated pipelines.
• Zero Dependencies: Both scripts rely exclusively on Python standard libraries, ensuring portability across any CI environment or local development machine.
• Claude Integration: The SKILL.md wrapper enables seamless usage through Claude Code's plugin ecosystem with standardized command interfaces.

Frequently Asked Questions

How does the stack detector determine which package manager to use for Node.js projects?

The detector analyzes lock file presence to select the appropriate installer command. When it identifies package-lock.json, it recommends npm ci; for pnpm-lock.yaml, it selects pnpm install --frozen-lockfile; and for yarn.lock, it chooses yarn install --frozen-lockfile. This logic ensures reproducible builds by respecting the specific package manager used in the original project.

Can I customize the generated pipeline templates?

Yes, while the generator uses reference templates stored in references/github-actions-templates.md and references/gitlab-ci-templates.md, you can modify the Python source in pipeline_generator.py to adjust job sequences, add custom steps, or change base images. The modular design separates template strings from detection logic for straightforward customization.

Does this tool support monorepo structures with multiple languages?

The StackReport dataclass aggregates all detected signals within a repository scan, enabling generation of multi-job workflows that handle different languages simultaneously. For example, a repository containing both package.json and pyproject.toml will generate distinct jobs for Node.js and Python testing within the same CI file.

Is internet connectivity required during pipeline generation?

No, both stack_detector.py and pipeline_generator.py operate entirely offline using local file system analysis and embedded template strings. The zero-dependency, standard-library-only architecture ensures functionality in air-gapped environments or secure build pipelines without external package downloads.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}