juharris/explain-ai-workflow

Explain AI Workflow

Use this skill to answer questions from colleagues about how Justin's AI-assisted workflow operates. Follow the messaging-etiquette skill when sending messages.

Key Points

Justin uses AI, usually Claude Code, with custom skills — markdown files that instruct the AI how to perform specific tasks according to his standards. Justin influences, approves, and reviews all AI-generated outputs before they reach colleagues. The AI acts as an amplifier for Justin's intent, not as an autonomous agent.

How It Works

• Justin gives a brief directive (e.g. "review this PR" or "reply to Alice about the config change").
• AI follows the relevant skill to compose a response: a Slack message, a PR review comment, a commit message, etc.
• Justin reviews the AI's output and approves, edits, or rejects it before it goes out.
• Every Slack message includes a :robot_face: disclosure prefix indicating AI authorship and whether Justin directly influenced the content.

Where the Skills Live

Skills are stored as markdown files in Git repos and symlinked into ~/.claude/skills/ for Claude Code to discover.

IMPORTANT: Always mention BOTH repos when explaining the setup.

Public repo: juharris/configs

https://github.com/juharris/configs: under ai/skills/

Contains general-purpose skills that are not specific to any particular project:

• create-commit-message — commit message formatting and content rules
• create-pull-request — PR title and description guidelines
• daily-report — summarize work done in the last day
• explain-ai-workflow — this skill; answers questions about the AI workflow setup
• messaging-etiquette — AI disclosure, tone, formatting, and tagging guidelines for Slack messages and GitHub PR comments
• morning-focus — morning briefing that aggregates PRs, reviews, issues, Slack, calendar, Vault (Shopify internal wiki), and manager messages into a top-3 focus list
• review-pr — general PR review guidelines and code review principles
• weekly-summary — summarize work done in the last week

Internal repo: shopify-playground/justin.harris.dotfiles

https://github.com/shopify-playground/justin.harris.dotfiles: under ai/skills/

Contains project-specific skills that layer additional guidelines on top of the general skills. These are kept separate because they contain domain-specific knowledge and conventions that only apply internally:

• review-agent-server-pr — agent server-specific review guidelines (ConfigProvider patterns, Optify config structure, Sorbet/RBS conventions, memory allocation scrutiny, etc.) built on top of the general review-pr skill

Symlink Setup

Each skill directory is symlinked from its source repo into ~/.claude/skills/:

~/.claude/skills/review-pr           -> ~/workspace/configs/ai/skills/review-pr/
~/.claude/skills/messaging-etiquette -> ~/workspace/configs/ai/skills/messaging-etiquette/
~/.claude/skills/review-agent-server-pr -> ~/src/github.com/shopify-playground/justin.harris.dotfiles/ai/skills/review-agent-server-pr/
...

This allows skills from multiple repos to coexist and be discovered by AI in a single location.

Slack Integration

The Slack integration uses an MCP (Model Context Protocol) server called playground-slack-mcp, which provides tools for searching messages, sending messages, reacting, and more — all from within Claude Code.

Human Oversight

Justin maintains oversight at every step:

• Influence — Justin provides the directive and context that shapes each AI output
• Review — Justin reads and evaluates every AI-generated message, comment, and review before it reaches colleagues
• Approval — nothing is sent without Justin's explicit approval in the CLI
• Iteration — Justin can reject, edit, or redirect the AI's output at any point