cyrus-setup（赛勒斯设置）

Cyrus Setup

One-command setup for self-hosted Cyrus. This orchestrator walks you through everything needed to run Claude Code as a background agent from Linear, Slack, and GitHub.

CRITICAL Rules

Never Read or Write ~/.cyrus/.env Directly

FORBIDDEN: Do NOT use Read, Edit, or Write tools on ~/.cyrus/.env or any file inside ~/.cyrus/. This file contains secrets (API keys, tokens, signing secrets). All interaction with this file MUST go through Bash commands (grep, printf >>..., etc.) which the user can see and approve. Never read its contents into the conversation context.

Browser Automation

The goal of browser automation in this skill is to reduce sign-in and setup fatigue — the agent navigates web UIs, fills forms, and scrapes credentials so the user doesn't have to do it all manually.

Three modes, in order of preference:

1. claude-in-chrome (preferred when available) — if the user is running Claude Code and has the claude-in-chrome MCP extension connected, use it. This has the huge advantage of using the user's existing Chrome with all their signed-in sessions (Linear, Slack, GitHub). Check availability by seeing if mcp__claude-in-chrome__* tools exist.
2. agent-browser CLI — a standalone Playwright-based binary invoked via Bash. Requires launching a fresh Chrome profile with remote debugging enabled (the user will need to sign in to services in that profile). Check with which agent-browser.
3. Manual guided flow — the user follows the agent's step-by-step instructions and does the clicks themselves. Always available as Path B in each sub-skill.

Determining which mode to use:

1. First, check if claude-in-chrome MCP tools are available in the current session. If yes, use those — no setup needed.
2. If not, check which agent-browser. If installed, launch a fresh Chrome profile (see below).
3. If neither is available, ask the user: install agent-browser, or follow manual instructions?

agent-browser setup — fresh Chrome profile:

agent-browser needs a Chrome instance with remote debugging enabled. Launch one with an isolated profile:

# Find an open port
for port in $(seq 9222 9322); do
  if ! lsof -i :"$port" > /dev/null 2>&1; then
    echo "Open port found: $port"
    break
  fi
done

# Create a fresh profile directory
mkdir -p ~/.cyrus/chrome-profile

# Launch Chrome with remote debugging (runs in background)
# macOS:
"/Applications/Google Chrome.app/Contents/MacOS/Google Chrome" \
  --remote-debugging-port=<port> \
  --user-data-dir="$HOME/.cyrus/chrome-profile" &

# Linux:
google-chrome --remote-debugging-port=<port> \
  --user-data-dir="$HOME/.cyrus/chrome-profile" &

# Connect agent-browser to it
agent-browser connect <port>

After connecting, commands work normally:

agent-browser navigate "https://example.com"
agent-browser click "button:text('Submit')"
agent-browser fill "#input-id" "value"
agent-browser screenshot
agent-browser eval "document.title"

Important: The user will need to sign in to Linear, Slack, and GitHub in this fresh Chrome profile before the agent can automate app creation. The agent should navigate to each service and pause for the user to sign in before proceeding with automation.

Cleanup: After setup is complete, close the Chrome instance:

lsof -ti :<port> | xargs kill 2>/dev/null

How This Works

This skill runs sub-skills in order, skipping any that are already complete. You can re-run /cyrus-setup at any time to add integrations or fix configuration.

Loading Sub-Skills

Each step references a sub-skill file. To execute a sub-skill, read the SKILL.md file using the Read tool and follow its instructions. The sub-skill files are sibling directories to this skill:

To find the files, look for them relative to this file's directory (go up one level, then into the sub-skill directory). For example, if this file is at ~/.claude/skills/cyrus-setup/SKILL.md, the sub-skills are at ~/.claude/skills/cyrus-setup-prerequisites/SKILL.md, etc.

If a sub-skill file is not found, use Glob to search for it: **/cyrus-setup-prerequisites/SKILL.md

Step 0: Identity & Surface Selection

Before anything else, collect preferences from the user. Use the AskUserQuestion tool if available — ask questions interactively rather than printing them as a text block. You may bundle related questions into a single ask, or ask them one at a time.

Question 1: Name & Description

Ask the user (defaults in parentheses):

• What would you like to name your agent? — This name appears in Linear, Slack, and GitHub integrations. (default: Cyrus)
• Give your agent a short description — one sentence, shown in integration app listings. (default: AI coding agent for automated development)

Store as AGENT_NAME and AGENT_DESCRIPTION — used when creating Linear, Slack, and GitHub apps.

Question 2: Which surfaces?

Ask the user to select one or more:

• Linear — issue tracking, recommended for most users
• GitHub — PR comments and issues
• GitLab — MR comments and issues
• Slack — chat messages

At least one is required. Store the selection — it determines which integration sub-skills run (Steps 4-6).

Question 3: Package manager?

Ask: npm, pnpm, bun, or yarn?

Store the answer — used by the prerequisites skill.

Step 1: Prerequisites

Read the cyrus-setup-prerequisites/SKILL.md sub-skill and follow its instructions.

Pass the user's package manager preference from Step 0.

Step 2: Claude Auth

Read the cyrus-setup-claude-auth/SKILL.md sub-skill and follow its instructions.

Step 3: Webhook Endpoint

Read the cyrus-setup-endpoint/SKILL.md sub-skill and follow its instructions.

Step 4: Linear (if selected)

Only if the user selected Linear in Step 0.

Read the cyrus-setup-linear/SKILL.md sub-skill and follow its instructions.

Step 5: GitHub (if selected)

Only if the user selected GitHub in Step 0.

Read the cyrus-setup-github/SKILL.md sub-skill and follow its instructions.

Step 5b: GitLab (if selected)

Only if the user selected GitLab in Step 0.

Read the cyrus-setup-gitlab/SKILL.md sub-skill and follow its instructions.

Step 6: Slack (if selected)

Only if the user selected Slack in Step 0.

Read the cyrus-setup-slack/SKILL.md sub-skill and follow its instructions. All paths (A-1, A-2, and B) must use the manifest-based creation flow — never create the Slack app "from scratch" with manual scope/event configuration.

Step 7: Add Repositories

Read the cyrus-setup-repository/SKILL.md sub-skill and follow its instructions.

Step 8: Launch

Read the cyrus-setup-launch/SKILL.md sub-skill and follow its instructions.

Design Principles

1. Skip-if-done — Every sub-skill checks existing state first. Re-running /setup is safe.
2. Secrets never enter chat — Credentials are either scraped via agent-browser or written via clipboard-to-env shell commands the user runs in their terminal.
3. Agent writes non-secret config — Values like CYRUS_SERVER_PORT and LINEAR_DIRECT_WEBHOOKS are written directly by the agent.
4. Browser automation when available — Uses agent-browser for Linear/Slack app creation; falls back to guided manual steps if not installed.
5. Package manager aware — The user's choice is used consistently throughout.