Update from External Sources

Your team’s coding knowledge doesn’t live only in the codebase. Valuable conventions, decisions, and best practices are scattered across pull request reviews, Slack discussions, issue trackers, and documentation wikis. Packmind can automatically mine these external sources using AI agent skills and MCP servers, classify findings for playbook relevance, and submit them as change proposals for your team to review.

Looking to do a one-time import from a knowledge base? See Import from Confluence, Notion and more.

Why Mine External Sources?

• Knowledge is scattered — PR review comments, Slack threads, Jira issues, and wiki pages all contain insights that should live in the playbook.
• Manual updates are easy to forget — Without automation, new conventions agreed upon in a PR review or a Slack discussion may never make it into the playbook.
• Continuous mining keeps the playbook current — Regularly scanning external sources ensures emerging patterns and decisions are captured as they happen.
• Change proposals ensure human review — Every finding goes through a change proposal, so your team stays in control of what enters the playbook.

How It Works

The workflow follows a three-stage pipeline:

1. Fetch — A source skill queries an external tool through its MCP server (e.g., fetching merged PR review comments from GitHub, or retrieving recent channel messages from Slack).
2. Classify — The AI agent analyzes the fetched data and identifies items relevant to coding conventions, best practices, architectural decisions, or recurring patterns.
3. Update — Classified findings are handed off to the packmind-update-playbook skill, which creates change proposals in Packmind — standards, commands, or skills depending on the nature of each finding.

This workflow is AI agent agnostic. It works with any agent that supports MCP — Claude Code, GitHub Copilot, Cursor, Windsurf, and others.

Supported Sources

Source	What is mined	MCP Server
GitHub PR Comments	Review comments from merged pull requests	GitHub MCP
Slack Conversations	Channel discussions and threaded messages	Slack MCP
Jira Issues	Resolved issues and their comments	Jira MCP
GitLab MR Comments	Review comments from merged merge requests	GitLab MCP
Confluence Documentation	Pages and spaces	Atlassian MCP
Notion Documentation	Pages and databases	Notion MCP

These are example use cases. Any tool that exposes an MCP server can be used as a source — see Adding Custom Sources below.

Example Skills Repository

The demo-use-case-skills repository contains ready-to-use source skills for all six integrations listed above. Inside you will find:

• Source skills for each integration — one skill per external source, each handling data fetching, noise filtering, and relevance classification.
• Shared skills — packmind-update-playbook (creates change proposals from findings) and packmind-cli-list-commands (prevents duplicate artifacts).
• CI workflow examples — GitHub Actions configurations for running source skills on a schedule, so your playbook is updated automatically without manual intervention.

Each integration folder includes its own README with setup instructions, MCP server configuration, and usage examples. Head to the repository to get started.

Adding Custom Sources

The architecture is extensible by design. To add a new external source, create a skill that fetches data from the source’s MCP server, classifies findings for playbook relevance, and hands them off to packmind-update-playbook. Any tool with an MCP server can become a source — monitoring tools, design systems, internal APIs, and more.

Learn More

• Change Proposals — How change proposals are reviewed and applied
• Updating Your Playbook — Overview of all playbook update workflows
• Import from Confluence, Notion and more — One-time import from knowledge bases
• Skills Management — How skills work and how to create them
• CLI Reference — CLI installation and commands
• Demo Use Case Skills — Example source skills repository