‍It's Wednesday afternoon, you're about to deploy a major release, and someone asks, "Where are the release notes?" Your heart sinks as you realize you'll be spending the next three hours combing through commit messages, trying to translate technical jargon into user-friendly language, and inevitably missing half the important changes.

At Ascend, we empower data teams to build fast, optimized, and automated data pipelines with AI-integrated workflows. We practice what we preach by applying the same automation principles to our own release notes. After searching for articles on a complete "full-stack" release notes pipeline and coming up empty-handed, I decided to build one myself—and I want to share how you can too.

In this post, you'll learn how to transform your release process from a manual, error-prone chore into a smooth, AI-powered automation that your team will actually love. We'll cover everything from architecture decisions to implementation gotchas, plus the real-world impact on our development workflow.

The Problem: Why Manual Release Notes Fail

Before diving into solutions, let's acknowledge why manual release notes are such a persistent pain point for development teams.

Time sink that derails productivity. Technical writers spend hours per release cycle syncing with developers and hand-writing release notes. This isn't just about the time—it's about context switching at the worst possible moment, right when you're trying to ship.

Procrastination leads to rushed or missing documentation. Release notes often become an afterthought, hastily written during deployment or skipped entirely when deadlines loom. The result? Users discover new features by accident, and support teams field confused questions about undocumented changes.

Multi-repository complexity multiplies the pain. When your product spans multiple repositories with independent release cycles, coordinating release notes becomes a logistical nightmare. Changes get missed, duplicated, or attributed to the wrong release.

Technical commits don't translate to user benefits. Raw commit messages like `fix null pointer exception in data validation` don't help users understand that their data processing is now more reliable. The translation from technical change to user value requires domain knowledge and communication skills that not every developer possesses.

The result? Inconsistent, incomplete, or entirely missing release notes that frustrate users and waste everyone's time.

‍

Requirements and Prerequisites

To build your own automated release notes pipeline using our framework, you'll need the following setup:

Essential Access and Permissions

- GitHub account with admin access to target repositories

- GitHub CLI authentication configured

- Repository permissions for creating branches, commits, and pull requests

Security and Authentication

- GitHub App with the following permissions on target repositories:

•  Contents: Read and Write (for accessing commit history and updating files)
•  Pull Requests: Read and Write (for creating automated PRs)‍
• Metadata: Read (for repository information)

- OpenAI API key with sufficient credits for text generation

- OpenAI API key and GitHub App private key stored as secrets in your target repository

Technical Environment

- Python environment

- Basic familiarity with GitHub Actions workflows

- Understanding of Git commit history and repository structure

Pro tip: Set up the GitHub App first—it provides more secure, granular permissions than personal access tokens and works better in automated environments. The setup takes 10 minutes but saves hours of authentication headaches later.

‍

Architecture Overview: How It All Works Together

Our pipeline follows a simple but powerful workflow designed for flexibility and transparency:

Input Flexibility

The system accepts two types of input to accommodate different release strategies:

• Date-based approach: Specify a "since" date to capture all commits across repositories from that point forward‍
• Version-specific approach: Provide a JSON string mapping specific repository versions for more granular control

This dual approach means you can generate weekly summaries ("show me everything since last Friday") or targeted release notes ("show me changes in repo A since v2.1.0 and repo B since v1.5.2").

‍

Core Processing Pipeline

1. Commit Harvesting: GitHub Actions trigger Python scripts that fetch repository commit histories from their `main` branches based on your input specifications.

2. AI Summarization: Leverage the OpenAI API to process raw commit messages and transform them into user-friendly release notes with three key sections:

• Features: Major functionality additions that users will notice
• Improvements: Enhancements to existing features, performance optimizations
• Bug Fixes: Resolved issues and stability improvements

3. Content Integration: The system automatically:

•   Extracts the "week of" date for consistent release note headers
•   Prepends the AI-generated summary to the existing "What's new" documentation

4. Pull Request Creation: The pipeline creates a well-formatted PR in your documentation repository that includes:

•    Updated "What's new" page with the new release notes
•    Complete raw commit history in the PR description for team review
•    Proper formatting and consistent styling

Why This Architecture Works

• Separation of concerns: Data collection, AI processing, and documentation updates are distinct steps, making debugging and maintenance easier
  ‍
• Transparency: Raw commit data is preserved alongside the AI summary, so reviewers can verify accuracy
  ‍
• Review process: PRs allow for human oversight before publication, catching edge cases and hallucinations and ensuring quality
  ‍
• Flexibility: Supports both scheduled releases and ad-hoc updates without changing the core workflow

Strategy: Choosing the Right Approach for Your Team

Our multi-repository challenge shaped our solution, but your team's structure should drive your approach.

Why We Chose Commit-Based Analysis

At Ascend, we manage several product repositories, each on independent versioning schedules. Traditional release management tools didn't fit our distributed development model, so we opted to analyze commit history directly.

Benefits of our approach:

• Real-time accuracy: Captures all changes as they happen, not just tagged releases
• Cross-repository coordination: Aggregates changes from multiple repos into unified release notes
• Granular control: Can generate release notes for any time period or specific version ranges
• No additional tooling: Works with standard Git workflows without requiring release management overhead

‍

Alternative Strategies to Consider

‍If your team uses unified release management:

• Look into tools like Release Please, Semantic Release, or Changesets with descriptive release information
• Consider AI-summarizing these structured releases instead of raw commits

This approach works well for monorepos or teams with standardized release processes

Hybrid approaches:

• Combine commit analysis with existing release tools for comprehensive coverage
• Use commit data to fill gaps between formal releases
• Supplement automated generation with manual curation for major releases

Choose based on your team's workflow, repository structure, and existing tooling. There's no one-size-fits-all solution, but the principles we'll cover adapt to most scenarios.

‍

Source Code Deep Dive: The Python Engine

The heart of our pipeline is Python functionality. Here's how the key components work:

Commit Data Collection

def fetch_commit_history(
  repos: Union[str, List[str], pathlib.Path],
  timeout_seconds: int = 120,
  since_date: Optional[str] = None,
  from_ref: Optional[str] = None,
  to_ref: Optional[str] = None,
) -> Dict[str, List[Dict[str, Any]]]:
  """
  Fetches commit history from one or multiple GitHub repositories using the GitHub CLI.
  Works with both public and private repositories, provided the authenticated user has access.
  """
  # Check GitHub CLI is installed
  subprocess.run(
    ["gh", "--version"],
    capture_output=True,
    check=True,
    timeout=timeout_seconds,
  )
  # Process the repos input to handle various formats
  if isinstance(repos, pathlib.Path) or (isinstance(repos, str) and os.path.exists(repos) and repos.endswith(".json")):
    with open(repos, "r") as f:
      repos = json.load(f)
  elif isinstance(repos, str):
    repos = [repo.strip() for repo in repos.split(",")]
  results = {}
  for repo in repos:
    # Get repository info and default branch
    default_branch_cmd = subprocess.run(
      ["gh", "api", f"/repos/{repo}"],
      capture_output=True,
      text=True,
      check=True,
      timeout=timeout_seconds,
    )
    repo_info = json.loads(default_branch_cmd.stdout)
    default_branch = repo_info.get("default_branch", "main")
    # Build API query with parameters
    api_path = f"/repos/{repo}/commits"
    query_params = ["per_page=100"]
    if since_date:
      query_params.append(f"since={since_date}T00:00:00Z")
    target_ref = to_ref or default_branch
    query_params.append(f"sha={target_ref}")
    api_url = f"{api_path}?{'&'.join(query_params)}"
    # Fetch commits using GitHub CLI
    result = subprocess.run(
      ["gh", "api", api_url],
      capture_output=True,
      text=True,
      check=True,
      timeout=timeout_seconds,
    )
    commits = json.loads(result.stdout)
    results[repo] = commits
  return results

Key implementation details:

• GitHub CLI integration: Uses the `gh` command-line tool for authenticated API access to both public and private repositories
• Flexible input handling: Accepts single repos, comma-separated lists, or JSON files containing repository lists
• Robust error handling: Validates GitHub CLI installation and repository access before attempting to fetch commits
• Configurable date filtering: Supports both date-based and ref-based commit filtering

‍

AI-Powered Summarization

def summarize_text(content: str, api_key: Optional[str] = None) -> str:
  """
  Summarize provided text content (e.g., commit messages) using OpenAI API.
  """
  if not content.strip():
    return "No commit data found to summarize"
  # Get API key from parameter or environment
  api_key = api_key or os.getenv("OPENAI_API_KEY")
  if not api_key:
    raise RuntimeError("OpenAI API key not found. Set the OPENAI_API_KEY environment variable.")
  client = OpenAI(api_key=api_key)
  response = client.chat.completions.create(
    model="gpt-4o",
    messages=[
      {"role": "system", "content": SYSTEM_PROMPT},
      {"role": "user", "content": content},
    ],
    temperature=0.1,
    max_tokens=1000,
  )
  return response.choices[0].message.content.strip()

def summarize_commits(content: str, add_date_header: bool = True) -> str:
  """
  Summarize commit content and optionally add a date header.
  """
  summary_body = summarize_text(content)
  if add_date_header:
    # Add header with week date
    now_iso = datetime.utcnow().strftime("%Y-%m-%dT%H:%M:%SZ")
    monday = get_monday_of_week(now_iso)
    return f"## 🗓️ Week of {monday}\n\n{summary_body}"
  return summary_body

‍

Our initial system prompt for consistent categorization:

You are a commit message organizer. Analyze the commit messages and organize them into a clear summary.
Group similar commits and format as bullet points under these categories:
- 🚀 Features
- ⚠️ Breaking changes
- 🌟 Improvements
- 🛠️ Bug fixes
- 📝 Additional changes
...
Within the Improvements section, do not simply say "Improved X" or "Fixed Y" or "Added Z" or "Removed W".
Instead, provide a more detailed and user-relevant description of the improvement or fix.
Convert technical commit messages to user-friendly descriptions and remove PR numbers and other technical IDs.
Focus on changes that would be relevant to users and skip internal technical changes.
Format specifications:
- Format entries as bullet points: "- [Feature description]"
- Use clear, user-friendly language while preserving technical terms
- For each item, convert technical commit messages to user-friendly descriptions:
   - "add line" → "New line functionality has been added"
   - "fix css overflow" → "CSS overflow issue has been fixed"
- Capitalize Ascend-specific terms in bullet points such as "Components"
Strictly exclude the following from your output:
- Any mentions of branches (main, master, develop, feature, etc.)
- Any mentions of AI rules such as "Added the ability to specify keywords for rules"
- Any references to branch integration or merges
- Any language about "added to branch" or "integrated into branch"
- Dependency upgrades and version bumps

Prompt engineering:

• Structured categorization: Our prompt enforces specific emoji-categorized sections for consistent output formatting
• User-focused translation: Explicitly instructs the AI to convert technical commits into user-friendly language
• Content filtering: Automatically excludes dependency updates, test changes, and internal technical modifications
• Low temperature setting: Uses 0.1 temperature for consistent, factual output rather than creative interpretation

‍

Content Integration and File Management

def get_monday_of_week(date_str: str) -> str:
  """
  Get the Monday of the week containing the given date, except for Sunday which returns the next Monday.
  """
  date = datetime.strptime(date_str, "%Y-%m-%dT%H:%M:%SZ")
  # For Sunday (weekday 6), get the following Monday
  if date.weekday() == 6:  # Sunday
    days_ahead = 1
  else:  # For all other days, get the Monday of the current week
    days_behind = date.weekday()
    days_ahead = -days_behind
  target_monday = date + timedelta(days=days_ahead)
  return target_monday.strftime("%Y-%m-%d")

File handling considerations:

• Consistent date formatting: Automatically calculates the Monday of the current week for consistent release note headers
• Encoding safety: Properly handles Unicode characters in commit messages from international contributors
• Atomic file operations: Uses temporary files during processing to prevent corruption if the process is interrupted

‍

GitHub Actions: Orchestrating the Automation

Our workflow ties everything together with robust automation that handles the complexities of CI/CD environments.

‍

Workflow Triggers and Inputs

name: Weekly Release Notes Update
on:
  workflow_dispatch:
    inputs:
      year:
        description: 'Year (YYYY) of date to start collecting releases from'
        default: '2025'
      month:
        description: 'Month (MM) of date to start collecting releases from'
        default: '01'
      day:
        description: 'Day (DD) of date to start collecting releases from'
        default: '01'
      repo_filters:
        description: 'JSON string defining filters for specific repos'
        required: false
      timeout_seconds:
        description: 'Timeout in seconds for API calls'
        default: '45'

Flexible triggering options:

• Manual dispatch with granular date control: Separate year, month, day inputs for precise date filtering
• Repository-specific filtering: JSON configuration allows different filtering strategies per repository
• Configurable timeouts: Adjustable API timeout settings for different network conditions

‍

Secure Authentication Flow

- uses: actions/create-github-app-token@v2
  id: app-token
  with:
    app-id: <YOUR-APP-ID>
    private-key: ${{ secrets.GHA_DOCS_PRIVATE_KEY }}
    owner: ascend-io
    repositories: ascend-docs,ascend-core,ascend-ui,ascend-backend

Security best practices:

• GitHub App with specific repository access: Explicitly lists only the repositories that need access
• Scoped permissions: App configured with minimal necessary permissions for the specific repositories
• Secret management: Private key stored securely in GitHub Secrets

‍

Repository Configuration Processing

- name: Prepare repository filter configuration
  run: |
    CONFIG_FILE=$(mktemp)
    echo "{}" > "$CONFIG_FILE"
    if [ -n "${{ github.event.inputs.repo_filters }}" ]; then
      echo '${{ github.event.inputs.repo_filters }}' > "$CONFIG_FILE"
    else
      DATE_STRING="${{ github.event.inputs.year }}-${{ github.event.inputs.month }}-${{ github.event.inputs.day }}"
      jq -r '.[]' bin/release_notes/input_repos.json | while read -r REPO; do
        FILTER="since:$DATE_STRING"
        jq --arg repo "$REPO" --arg filter "$FILTER" '. + {($repo): $filter}' "$CONFIG_FILE" > "${CONFIG_FILE}.tmp" && mv "${CONFIG_FILE}.tmp" "$CONFIG_FILE"
      done
    fi
    CONFIG_JSON=$(cat "$CONFIG_FILE")
    echo "config_json<<EOF" >> $GITHUB_OUTPUT
    echo "$CONFIG_JSON" >> $GITHUB_OUTPUT
    echo "EOF" >> $GITHUB_OUTPUT

‍

Data Processing and File Management

- name: Generate release notes
  env:
    OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
    GITHUB_TOKEN: ${{ steps.app-token.outputs.token }}
  run: |
    CONFIG_JSON='${{ steps.repo_config.outputs.config_json }}'
    CONFIG_FILE=$(mktemp)
    echo "$CONFIG_JSON" > "$CONFIG_FILE"
    RAW_OUTPUT=$(python bin/release_notes/generate_release_notes.py \
      --repo-config-string "$(cat "$CONFIG_FILE")" \
      --timeout "${{ github.event.inputs.timeout_seconds }}")
    # Split summary and commits using delimiter
    SUMMARY=$(echo "$RAW_OUTPUT" | sed -n '1,/^### END SUMMARY ###$/p' | sed '$d')
    MONDAY_DATE=$(echo "$SUMMARY" | head -n1 | grep -oE "[0-9]{4}-[0-9]{2}-[0-9]{2}")
    echo "monday_date=$MONDAY_DATE" >> $GITHUB_OUTPUT
    echo 'summary<<EOF' >> $GITHUB_OUTPUT
    echo "$SUMMARY" >> $GITHUB_OUTPUT
    echo 'EOF' >> $GITHUB_OUTPUT

Key implementation lessons:

• Temporary file strategy: We learned the hard way that GitHub Actions environments can lose data between steps. Writing to temporary files solved reliability issues where data would appear blank in subsequent steps.
• Complex JSON handling: Uses `jq` for safe JSON manipulation and temporary files to avoid shell quoting issues with complex JSON strings
• Output parsing: Logic to split AI-generated summaries from raw commit data using delimiter markers
• Robust error handling: `set -euo pipefail` ensures the script fails fast on any error, preventing silent failures

‍

File Integration and Pull Request Creation

- name: Update whats-new.mdx with release notes
  run: |
    FILE="website/docs/whats-new.mdx"
    BRANCH_NAME="notes-${{ steps.generate_notes.outputs.monday_date }}"
    git branch $BRANCH_NAME main
    git switch $BRANCH_NAME
    TEMP_SUMMARY_FILE=$(mktemp)
    echo '${{ steps.generate_notes.outputs.summary }}' > "$TEMP_SUMMARY_FILE"
    cat "$TEMP_SUMMARY_FILE" "$FILE" > "${FILE}.new"
    mv "${FILE}.new" "$FILE"
    rm -f "$TEMP_SUMMARY_FILE"

File management features:

• Atomic file operations: Uses temporary files and atomic moves to prevent file corruption
• Branch management: Creates date-based branches for organized PR tracking
• Content preservation: Carefully prepends new content while preserving existing documentation structure

Lessons Learned and Best Practices

Building this pipeline taught us valuable lessons about documentation automation that go beyond the technical implementation.

‍

Technical Insights

File persistence matters in CI/CD environments. GitHub Actions environments can be unpredictable—always write important data to files rather than relying on environment variables or memory. We learned this the hard way when release notes would mysteriously appear blank in PRs.

‍

API reliability requires defensive programming. Build retry logic and fallbacks for external API calls (OpenAI, GitHub). Network issues and rate limits are inevitable, especially as your usage scales.

‍

Prompt engineering is crucial for consistent output. Spend time crafting prompts that consistently produce the format and tone you want. Small changes in wording can dramatically affect AI output quality and consistency.

‍

Human review is essential, even with AI generation. Having team members review PRs catches edge cases, ensures quality, and builds confidence in the automated system. The goal isn't to eliminate human oversight—it's to make it more efficient and focused.

‍

Historical tracking and product evolution insights. Automated generation creates a consistent record of product evolution that's valuable for retrospectives, planning, and onboarding new team members.

‍

Results and Impact

The automation has fundamentally transformed our release process and team dynamics:

‍

Quantifiable Improvements

Dramatic time savings: Reduced release note creation from 2-3 hours of writing time to 15 minutes of review time. That's a 90% reduction in effort while improving quality and consistency.

‍

Perfect consistency: Every release now has properly formatted, comprehensive notes. No more missed releases or inconsistent formatting across different team members.

‍

Increased frequency: We can now generate release notes weekly, providing users with more timely updates about product improvements.

‍

‍Complete coverage: Captures changes across all repositories without manual coordination, eliminating the risk of missing important updates.

‍

Next Steps and Future Enhancements

We're continuously improving the pipeline based on team feedback and evolving needs:

‍

Immediate Roadmap

Slack integration: Building a Slackbot to automatically share release notes with our community channels, extending the reach beyond just documentation updates.

‍

Repository tracing: Categorize the raw commits by repository and add links so it's easy to (literally) double-click into each PR for additional context.

‍

Future Possibilities

Multi-language support: Generating release notes in different languages for global audiences as we expand internationally.

‍

Ready to automate your own release notes? Start with the requirements above and build incrementally. Begin with a single repository, get the basic workflow running, then expand to multiple repos and add advanced features. Your future self (and your team) will thank you for eliminating this manual drudgery and creating a more consistent, professional release process.

The investment in automation pays dividends immediately—not just in time saved, but in the improved quality and consistency of your user communication. In a time where software moves fast, automated release notes ensure your documentation keeps pace.

‍

‍

Key implementation details:

• GitHub CLI integration: Uses the `gh` command-line tool for authenticated API access to both public and private repositories
  ‍
• Flexible input handling: Accepts single repos, comma-separated lists, or JSON files containing repository lists
  ‍
• Robust error handling: Validates GitHub CLI installation and repository access before attempting to fetch commits
  ‍
• Configurable date filtering: Supports both date-based and ref-based commit filtering

‍

AI-Powered Summarization

Our initial system prompt for consistent categorization:

Prompt engineering:

• Structured categorization: Our prompt enforces specific emoji-categorized sections for consistent output formatting
  ‍
• User-focused translation: Explicitly instructs the AI to convert technical commits into user-friendly language
  ‍
• Content filtering: Automatically excludes dependency updates, test changes, and internal technical modifications
  ‍
• Low temperature setting: Uses 0.1 temperature for consistent, factual output rather than creative interpretation

Content Integration and File Management

File handling considerations:

• Consistent date formatting: Automatically calculates the Monday of the current week for consistent release note headers
  ‍
• Encoding safety: Properly handles Unicode characters in commit messages from international contributors
  ‍
• Atomic file operations: Uses temporary files during processing to prevent corruption if the process is interrupted

GitHub Actions: Orchestrating the Automation

Our workflow ties everything together with robust automation that handles the complexities of CI/CD environments.

Workflow Triggers and Inputs

Flexible triggering options:

• Manual dispatch with granular date control: Separate year, month, day inputs for precise date filtering
  ‍
• Repository-specific filtering: JSON configuration allows different filtering strategies per repository
  ‍
• Configurable timeouts: Adjustable API timeout settings for different network conditions

Secure Authentication Flow

Security best practices:

• GitHub App with specific repository access: Explicitly lists only the repositories that need access
  ‍
• Scoped permissions: App configured with minimal necessary permissions for the specific repositories
  ‍
• Secret management: Private key stored securely in GitHub Secrets

Repository Configuration Processing

Data Processing and File Management

Key implementation lessons:

• Temporary file strategy: We learned the hard way that GitHub Actions environments can lose data between steps. Writing to temporary files solved reliability issues where data would appear blank in subsequent steps.
  ‍
• Complex JSON handling: Uses `jq` for safe JSON manipulation and temporary files to avoid shell quoting issues with complex JSON strings
  ‍
• Output parsing: Logic to split AI-generated summaries from raw commit data using delimiter markers
  ‍
• Robust error handling: `set -euo pipefail` ensures the script fails fast on any error, preventing silent failures

File Integration and Pull Request Creation

File management features:

• Atomic file operations: Uses temporary files and atomic moves to prevent file corruption
  ‍
• Branch management: Creates date-based branches for organized PR tracking
  ‍
• Content preservation: Carefully prepends new content while preserving existing documentation structure

Lessons Learned and Best Practices

Building this pipeline taught us valuable lessons about documentation automation that go beyond the technical implementation.

Technical Insights

File persistence matters in CI/CD environments. GitHub Actions environments can be unpredictable—always write important data to files rather than relying on environment variables or memory. We learned this the hard way when release notes would mysteriously appear blank in PRs.
‍

API reliability requires defensive programming. Build retry logic and fallbacks for external API calls (OpenAI, GitHub). Network issues and rate limits are inevitable, especially as your usage scales.
‍

Prompt engineering is crucial for consistent output. Spend time crafting prompts that consistently produce the format and tone you want. Small changes in wording can dramatically affect AI output quality and consistency.
‍

Human review is essential, even with AI generation. Having team members review PRs catches edge cases, ensures quality, and builds confidence in the automated system. The goal isn't to eliminate human oversight—it's to make it more efficient and focused.
‍

Historical tracking and product evolution insights. Automated generation creates a consistent record of product evolution that's valuable for retrospectives, planning, and onboarding new team members.
‍

Results and Impact

The automation has fundamentally transformed our release process and team dynamics:

Quantifiable Improvements

Dramatic time savings: Reduced release note creation from 2-3 hours of writing time to 15 minutes of review time. That's a 90% reduction in effort while improving quality and consistency.
‍

Perfect consistency: Every release now has properly formatted, comprehensive notes. No more missed releases or inconsistent formatting across different team members.

‍

Increased frequency: We can now generate release notes weekly, providing users with more timely updates about product improvements.

‍

Complete coverage: Captures changes across all repositories without manual coordination, eliminating the risk of missing important updates.

‍

Next Steps and Future Enhancements

We're continuously improving the pipeline based on team feedback and evolving needs:

Immediate Roadmap

Slack integration: Building a Slackbot to automatically share release notes with our community channels, extending the reach beyond just documentation updates.

‍

Repository tracing: Categorize the raw commits by repository and add links so it's easy to (literally) double-click into each PR for additional context.

Future Possibilities

Multi-language support: Generating release notes in different languages for global audiences as we expand internationally.
‍

Ready to automate your own release notes? Start with the requirements above and build incrementally. Begin with a single repository, get the basic workflow running, then expand to multiple repos and add advanced features. Your future self (and your team) will thank you for eliminating this manual drudgery and creating a more consistent, professional release process.

‍

The investment in automation pays dividends immediately—not just in time saved, but in the improved quality and consistency of your user communication. In a time where software moves fast, automated release notes ensure your documentation keeps pace.