# Claude Code Setup Instructions: Multica MVP Pod

Give this file to Claude Code running on the Mac/PC that has Multica CLI installed. Claude's job is to create/check the Multica workspace, four strong MVP agents, skills, assignments, and first SEO workflow request.

## Prerequisites

Before giving this file to Claude, install and authenticate the two required systems:

- Claude Code — local execution runtime for the agents.
  - Quickstart: https://docs.anthropic.com/en/docs/claude-code/quickstart
  - Setup docs: https://docs.anthropic.com/en/docs/claude-code/setup
- Multica — workspace, issue board, agent/skill system, and local daemon/runtime connection.
  - Docs: https://www.multica.ai/docs
  - CLI reference: https://multica.ai/docs/cli

Claude can check whether these are installed, but it cannot magically install/login for the human if an interactive step is required. Tiny but tragic distinction.

## Goal

Set up a small but useful Multica MVP pod:

- Claude Code installed on this Mac/PC.
- Multica CLI installed on this Mac/PC.
- Multica daemon running on this Mac/PC.
- Multica Cloud workspace selected.
- Four Multica agents using the same local Claude Code runtime:
  - Chief of Staff
  - Researcher
  - SEO Consultant
  - Blog Writer
- Core skills for evidence labels, final review, and readable HTML reports/previews.
- One sample parent issue assigned to Chief of Staff.

Important: the first sample workflow is SEO/blog focused, but the agents should not be hardcoded to that example. They should be useful general-purpose role agents that can work in a larger team later.

## Safety rules

- Do not print tokens, auth headers, secrets, or custom environment values.
- Do not publish anything publicly.
- Do not spend money.
- Do not modify production websites.
- If a command fails because setup/login is required, stop and tell the human exactly which interactive step is needed.
- Prefer JSON output and save IDs to a local setup file.
- Verify CLI flags with `multica <command> --help` if a command fails.

## Files to use

### Setup sections

- https://ydy.au/templates/setup/sections/01-claude-setup.md
- https://ydy.au/templates/setup/sections/02-create-agents.md
- https://ydy.au/templates/setup/sections/03-agent-instructions.md
- https://ydy.au/templates/setup/sections/04-attach-skills.md
- https://ydy.au/templates/setup/sections/05-run-first-request.md
- https://ydy.au/templates/setup/sections/06-review-loop.md

### Agent instruction files

- https://ydy.au/templates/agents/chief-of-staff.md
- https://ydy.au/templates/agents/researcher.md
- https://ydy.au/templates/agents/seo-consultant.md
- https://ydy.au/templates/agents/blog-writer.md

### Core skill files

- https://ydy.au/templates/skills/chief-of-staff-final-review.md
- https://ydy.au/templates/skills/evidence-labels.md
- https://ydy.au/templates/skills/html-styled-report-preview.md
- https://ydy.au/templates/skills/html-blog-approval-preview.md

### Optional workflow recipe

- https://ydy.au/templates/skills/route-seo-audit-to-blogs.md

## Step 1 — Check local prerequisites

Run:

```bash
claude --version
multica version
multica auth status || true
multica workspace list --output json
multica daemon status
multica runtime list --output json
```

Expected:

- Claude Code responds.
- Multica CLI responds.
- The user is logged in or can run `multica setup`.
- A workspace is selected or available.
- The daemon is running.
- A Claude Code runtime is visible in `multica runtime list`.

If not connected, run the interactive setup:

```bash
multica setup
```

Then re-run daemon/runtime checks.

## Step 2 — Select workspace and runtime

List workspaces and runtimes:

```bash
multica workspace list --output json
multica runtime list --output json > /tmp/multica-runtimes.json
```

If the correct workspace is not selected, switch to it:

```bash
multica workspace switch <workspace-id>
```

Find the runtime whose provider/tool/name indicates Claude Code and save it:

```bash
export MULTICA_CLAUDE_RUNTIME_ID="<runtime-id>"
```

If no Claude runtime exists, confirm `claude --version` works from the same shell, restart the daemon, and check again.

## Step 3 — Fetch starter files locally

```bash
mkdir -p /tmp/multica-mvp/{agents,skills,setup/sections}

for file in chief-of-staff researcher seo-consultant blog-writer; do
  curl -fsSL "https://ydy.au/templates/agents/$file.md" -o "/tmp/multica-mvp/agents/$file.md"
done

for file in chief-of-staff-final-review evidence-labels html-styled-report-preview html-blog-approval-preview; do
  curl -fsSL "https://ydy.au/templates/skills/$file.md" -o "/tmp/multica-mvp/skills/$file.md"
done

# Optional reference recipe for the sample SEO workflow. Do not create/attach this as a core skill unless you want the Chief of Staff to follow the example recipe very explicitly.
curl -fsSL "https://ydy.au/templates/skills/route-seo-audit-to-blogs.md" -o "/tmp/multica-mvp/skills/route-seo-audit-to-blogs.md"
```

## Step 4 — Create skills

Create these core skills in Multica, saving returned IDs:

- Chief of Staff Final Review
- Evidence Labels
- HTML Styled Report Preview
- HTML Blog Approval Preview

Do not create `Route SEO Audit to Site Growth Work` as a core skill by default. It is an optional workflow recipe/example. The Chief of Staff should route SEO work to the SEO Consultant because of the agent roles, not because a special routing skill exists.

Use the matching files in `/tmp/multica-mvp/skills/` as the skill content.

If the CLI supports JSON output, save IDs as variables like:

```bash
FINAL_REVIEW_SKILL_ID="<id>"
EVIDENCE_SKILL_ID="<id>"
HTML_REPORT_SKILL_ID="<id>"
HTML_BLOG_PREVIEW_SKILL_ID="<id>"
```

## Step 5 — Create the four MVP agents

Create each agent using the same Claude Code runtime and the matching instruction file:

1. Chief of Staff — `chief-of-staff.md`
2. Researcher — `researcher.md`
3. SEO Consultant — `seo-consultant.md`
4. Blog Writer — `blog-writer.md`

Save each returned agent ID.

## Step 6 — Assign skills

Suggested assignments:

- Chief of Staff: Chief of Staff Final Review, Evidence Labels, HTML Styled Report Preview, HTML Blog Approval Preview.
- Researcher: Evidence Labels, HTML Styled Report Preview.
- SEO Consultant: Evidence Labels, HTML Styled Report Preview, HTML Blog Approval Preview.
- Blog Writer: Evidence Labels, HTML Styled Report Preview, HTML Blog Approval Preview.

## Step 7 — Create the sample parent issue

Create one parent issue assigned to Chief of Staff:

```text
Do an SEO audit on https://example.com and have the team take actions to help us rank better.

Give me back a clear report on what you found, what the team did, and what needs approval before anything goes live.
```

The human should not need to tell the Chief of Staff how to create child tickets. That routing is the Chief of Staff's job.

## Step 8 — Verify the loop

Expected behaviour:

- Parent issue is assigned to Chief of Staff.
- Chief of Staff creates child tickets as needed.
- SEO Consultant audits the website.
- Researcher gathers supporting evidence if needed.
- Blog Writer creates blog content only if supported by the brief/findings.
- Substantial reports are readable HTML.
- Blog previews are created only if supported by the findings.
- Chief of Staff returns a final SEO/action report and approval list.

## If anything breaks

Report exactly:

- Command/action that failed.
- Exit/error message.
- Whether setup/login/workspace/runtime/agent/skill/issue creation is the blocker.
- The next human action required.

Do not silently invent successful setup. A false green tick is worse than a clear blocker.