Claude EM

Claude EM turns Claude into an Engineering Manager operating system.

Not a generic AI. Not a blank prompt.

A context-aware, opinionated system that understands:

• your team
• your context
• how you run engineering

Why Claude EM

It comes with:

1. Persistent context → your team + your way of working
2. Reusable workflows → skills that support real EM tasks

Claude EM is built for day-to-day EM work:

• Plan initiatives with clear scope and trade-offs
• Write epics and stories that engineers can actually build
• Analyze IC performance using real data (Jira + GitHub)
• Prepare 1:1s with context and intent
• Think through org and leadership decisions

It is intentionally CLI-first: faster, more predictable, and more cost-efficient (tokens).

Works out of the box with Jira and GitHub. Other tools can be added via CLI scripts or MCP servers.

---

Quick start (2 minutes)

1. Click on Use this template and create your own repository

2. Install Claude Code:

npm install -g @anthropic-ai/claude-code

3. Add your EM style and team context:

• Copy data/em_style_example.md → data/em_style.md and fill in your style
• Copy data/team_example.md → data/team_myteam.md and fill in your team and members

4. Open Claude in your workspace:

claude

5. Try your first prompt:

I want to write a discovery epic to assess frontend performance issues

6. Complete the full setup to unlock data-driven workflows (Jira, GitHub, and more)

---

Example use cases

• Plan an initiative

  We have recurring CI flakiness slowing down delivery. Plan an initiative to reduce it

• Analyze IC activity

  Analyze Maria’s last 30 days

• Prepare a 1:1

  Prepare a 1:1 with Ana. She’s delivering less lately and seems disengaged. I want to understand what’s going on

• Write a strategy

  Write a strategy to reduce operational load on the team while maintaining reliability

• Mentor

  I’m considering reorganizing the team into two squads. Help me think through trade-offs and risks

• Prepare the daily meeting

  Give me a daily-ready view of the team’s board

• Generate a user story map

  Generate a user story map from this PRD: [paste PRD]

• Check an epic

  Check this epic

• Write a delivery epic

  Write an epic to implement rate limiting in our API. Include scope, risks, and definition of done

• Write a discovery epic

  We have performance issues in the frontend. Write a discovery epic to understand root causes and options

• Write a weekly update

  Write the weekly update for the platform team Write the project update for the migration initiative

---

How it works

• CLAUDE.md defines how Claude should behave
• EM style (data/em_style.md) defines your personal way of working
• Skills (.claude/skills/) are reusable prompts for real EM tasks
• Team context (data/team_*.md) defines your team
• Data (data/) stores shared information (Jira, GitHub, etc.)
• Initiatives (initiatives/) are folders where work, analysis, and outputs live
• CLI-first approach: prefers CLI tools over MCP servers to stay fast, predictable, and low-cost in terms of tokens

---

Skills

---

Workspace structure

claude-em/
├── data/                       # Shared data across initiatives
│   ├── em_style.md             # Engineering Manager style
│   ├── team_{name}.md          # Team context files
│   ├── [source]/               # One folder per data source (jira, github, etc.)
│   └── tmp/                    # Temporary files
└── initiatives/                # All initiatives live here
    └── {initiative-name}/      # One folder per initiative
        ├── data/               # Initiative-specific data
        ├── tmp/                # Temporary files
        ├── scripts/            # Analysis and processing
        └── output/             # Results and reports

---

Setup (full)

1. Create your workspace

Click on Use this template and create your own repository.

This gives you a fully independent workspace you can customize.

---

2. Install Claude Code

npm install -g @anthropic-ai/claude-code

Docs: https://docs.anthropic.com/claude-code

---

3. Configure your EM profile

Define your EM style:

• Copy data/em_style_example.md → data/em_style.md
• Fill in your style and how you want Claude to respond

Then create your team context:

• Copy data/team_example.md
• Rename it to data/team_{your-team}.md
• Fill in your team context and members

---

4. (Optional) Install CLI tools

Claude EM is designed to work primarily with CLI tools instead of MCP, as they are faster and more cost-efficient.

Jira CLI (jira)

brew install ankitpokhrel/jira-cli/jira
jira init

Docs: https://github.com/ankitpokhrel/jira-cli

GitHub CLI (gh)

brew install gh
gh auth login

Docs: https://cli.github.com

---

5. (Optional) Configure credentials

Create a .env.local file at the workspace root (git-ignored):

# Jira
JIRA_URL="https://your-domain.atlassian.net"
JIRA_EMAIL="[email protected]"
JIRA_API_TOKEN="your-api-token"

# GitHub
GITHUB_TOKEN="your-github-token"

# Figma
FIGMA_TOKEN="your-figma-token"

Getting credentials:

• Jira API token: https://id.atlassian.com/manage-profile/security/api-tokens
• GitHub token: https://github.com/settings/tokens (scopes: repo, read:org, read:user)
• Figma token: https://www.figma.com/settings → Personal access tokens

---

6. (Optional) Configure the Jira skill for your project(s)

The /jira skill works with any Jira project, but you need to point it at yours. When you invoke it for the first time, Claude will ask for the missing values and offer to save them permanently.

If you manage more than one project, create one skill file per project:

# Copy the base skill for each project
cp .claude/skills/jira-project/SKILL.md .claude/skills/jira-myproject/SKILL.md
cp .claude/skills/jira-project/SKILL.md .claude/skills/jira-otherproject/SKILL.md

Then open each copy and update:

1. Folder name — already done by the cp command above
2. name: in the frontmatter — must match the folder name (e.g. jira-myproject)
3. description: in the frontmatter — replace “any Jira project” with your project name so Claude picks the right skill when you have multiple
4. The three configuration placeholders:

Each copy becomes its own skill. Claude will pick the right one based on context, or you can invoke them explicitly.

---

7. (Optional) Configure MCP servers

MCP servers give Claude direct access to Jira, GitHub, and Figma without CLI or scripts.

claude mcp add --plugin atlassian
claude mcp add --plugin github
claude mcp add --plugin figma

Slack (MCP only)

Slack has no practical CLI, so reading channels/threads is done through MCP. Add the official Slack connector and authenticate with your account:

claude mcp add --transport http slack https://mcp.slack.com/mcp

Note: Slack access depends on your channel/workspace permissions. DMs are not readable, and the history tools ignore date filters — Claude pulls recent messages and filters the window itself.

Tip: record your team’s key Slack channels in data/team_{name}.md so Claude knows where to look (e.g. an ## Slack section listing channel names). Otherwise it has to search and ask which channels apply.

---

8. (Optional) Get the most out of 1:1s

The /one-on-one skill works with nothing but a one-line prompt, but it gets sharper the more context you give it through the team file.

Per-person 1:1 doc. Give any member a living doc and the skill will read it when preparing the conversation — open topics, agreed actions, signals to watch, goals, and what happened in past 1:1s:

• Add a 1:1 file: link to the member in data/team_{your-team}.md
• The value can be a URL or a path, and the doc can be in any format — the skill adapts to whatever’s there
• For a suggested starting point, see .claude/skills/one-on-one/references/person-file-example.md
• It’s fully optional: if it’s missing or unreachable, the skill simply ignores it

Career Frameworks. Fill in the ## Career Frameworks section of the team file with links (per level) to your expectations docs. The skill reads the one matching the member’s seniority when the conversation turns to expectations or growth.

---

9. (Optional) Configure /write-update

The /write-update skill writes periodic team or project updates by pulling data from Jira, Slack, and configured URLs. It works out of the box with a one-line prompt, but gets significantly better with two things set up in the team file.

Templates. The skill ships with two built-in templates in .claude/skills/write-update/assets/:

You can create your own templates by adding .md files to that same assets/ folder. When you invoke the skill without specifying a template, Claude lists the available ones and asks you to choose. Each template can include an ## Instructions section to tell the skill which sources to query and how to scope the data.

Update Sources. Add an ## Update Sources section to data/team_{your-team}.md listing the URLs Claude should read — Confluence meeting notes, Atlas project pages, and so on. The skill fetches each URL in parallel and extracts relevant content for the date range.

Slack channels. Add a ## Slack section to the team file with the channel names to include in each update. Claude reads them in parallel and filters for signal (shipped work, decisions, risks, customer feedback).

---

Notes

• You don’t need full setup to start — Quick Start is enough
• CLI tools and MCP improve automation but are optional
• The system works best when your team context is accurate and up to date

---

Updating

Your workspace is yours.

Check the repository releases to see:

• New skills
• Improvements
• Bug fixes

Pull improvements when they’re useful.

---

Contributing

Contributions are welcome:

• Feedback from real usage
• Suggestions
• New skills
• Support for additional tools

---

License

MIT