Pushlog

A Docker-based GitHub Action that generates AI-powered changelogs from commits and code changes and posts them to Slack.

When a push occurs on a configured branch, the action:

1. Collects commit messages and metadata from the push event.
2. Extracts the code diff between the before/after commits.
3. Sends the data to the OpenAI API to generate a human-friendly changelog.
4. Posts the changelog to a Slack channel via webhook.

The changelog is written in plain language, understandable by both developers and non-technical stakeholders.

---

Required Secrets

Secret	Description
SLACK_WEBHOOK	Slack incoming webhook URL
OPENAI_API_KEY	OpenAI API key with access to GPT models

---

Inputs

Input	Required	Default	Description
slack_webhook	Yes	—	Slack webhook URL
openai_api_key	Yes	—	OpenAI API key
branch	Yes	—	Branch to monitor
language	No	en	Changelog language code (e.g. en, el, fr, de)
openai_model	No	gpt-4.1	OpenAI model to use for changelog generation
openai_temperature	No	0.1	Sampling temperature for changelog generation
openai_max_output_tokens	No	800	Maximum number of output tokens
slack_title	No	Repository Update	Title for the Slack message
slack_mentions	No	—	Slack mentions (e.g. <!channel>, <!here>, <@U012345>)
user_prompt	No	—	Freeform extra instructions for changelog generation
system_prompt	No	—	Optional system prompt override for the model behavior

Prompt Inputs

You can provide user_prompt as plain instructions.

Pushlog always sends the required context on its own:

• What the tool is doing
• Which language to use
• The commit summary
• The filtered code diff

That means users do not need placeholders or prompt templates. Whatever you put in user_prompt is treated as additional guidance layered on top of the default behavior.

You can also override system_prompt if you want to change the base model behavior. Pushlog still injects the selected language automatically.

OpenAI Settings

Pushlog now uses the OpenAI Responses API first and falls back to Chat Completions if needed.

• openai_model: controls quality, speed, and cost. The default is gpt-4.1 for stronger output quality.
• openai_temperature: lower values make the result more consistent. The default is 0.1.
• openai_max_output_tokens: caps the response size. The default is 800.

---

Example Workflow

name: Pushlog

on:
  push:
    branches:
      - main

jobs:
  changelog:
    runs-on: ubuntu-latest

    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0

      - uses: kmakris23/pushlog@v1
        with:
          slack_webhook: ${{ secrets.SLACK_WEBHOOK }}
          openai_api_key: ${{ secrets.OPENAI_API_KEY }}
          branch: main
          language: en
          openai_model: gpt-4.1
          slack_title: Repository Update
          slack_mentions: '<!channel>'

Example With User Prompt

name: Pushlog

on:
  push:
    branches:
      - main

jobs:
  changelog:
    runs-on: ubuntu-latest

    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0

      - uses: kmakris23/pushlog@v1
        with:
          slack_webhook: ${{ secrets.SLACK_WEBHOOK }}
          openai_api_key: ${{ secrets.OPENAI_API_KEY }}
          branch: main
          user_prompt: |
            Focus on user-facing impact first.
            Group the result into New Features, Fixes, and Maintenance.
            Keep each bullet short.

Example With System Prompt Override

name: Pushlog

on:
  push:
    branches:
      - main

jobs:
  changelog:
    runs-on: ubuntu-latest

    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0

      - uses: kmakris23/pushlog@v1
        with:
          slack_webhook: ${{ secrets.SLACK_WEBHOOK }}
          openai_api_key: ${{ secrets.OPENAI_API_KEY }}
          branch: main
          system_prompt: |
            You write release notes for engineering teams.
            Prefer crisp bullets and group related changes by theme.
            Focus on observable impact and skip low-value implementation details.
          user_prompt: |
            Keep the output compact.

Important: fetch-depth: 0 is required so the action can compute diffs across the full commit history.

---

Example Slack Output

@channel

Repository Update (main)

New Features
• Added login API with JWT authentication

Bug Fixes
• Fixed issue where duplicate orders could be created

Improvements
• Refactored invoice service for better maintainability

---

How It Works

1. Reads the GitHub push event payload from GITHUB_EVENT_PATH.
2. Extracts commit messages, authors, and file change metadata.
3. Runs git diff between the before and after SHAs.
4. Filters out noisy paths (node_modules, dist, build, lock files).
5. Truncates the diff to 15 000 characters to fit LLM context limits.
6. Sends the commit list and diff to OpenAI using the Responses API first, with a fallback to Chat Completions.
7. Posts the generated changelog to Slack.

---

License

MIT