agent-portfolio（Agent 组合）

Agent Portfolio — Introduced by My Agents

You are generating a personalized portfolio site from agent-reference reports. The site reflects the user's working style as observed by their AI collaborators — not a generic developer template, but a site shaped by real collaboration data. Each portfolio is different because each person's reports are different.

The result is an Astro site deployed to GitHub Pages at {username}.github.io, with sections for agent reviews, project highlights, work style visualization, and GitHub activity.

How This Skill Works

Most users will have already run the agent-reference skill to generate reports. If not, suggest running that skill first.

Step 1: Collect Inputs

This step gathers all data needed to generate the portfolio. Guide the user through each input:

1. Agent-reference reports (required)

• Ask the user where their report files are
• Expected structure: markdown files from agent-reference Phase 1 and optionally Phase 2
• If no reports exist, suggest running the agent-reference skill first and stop here

2. Resume / self-introduction (optional)

• Ask if they have a resume, CV, or self-intro to include
• This populates the "About" and "Career" sections
• Any format (markdown, PDF text, plain text)
• PDF handling: Read PDF files directly using your file reading tool (e.g., Claude Code's Read tool supports PDF natively) — do NOT use external conversion tools like pdftoppm or pdftotext, which lose structure and ordering. Even with direct reading, multi-column layouts, tables, and Korean text can be misinterpreted. Always verify the extracted content with the user — career order, company-project mapping, and tech stacks are commonly garbled. Markdown format is most reliable.
• PII warning: Resumes often contain phone numbers, email addresses, and home addresses. These must NOT appear on the public site. Extract only career/education/skills content. If the user wants a contact method on the site, suggest a contact form or LinkedIn link instead of raw PII.

3. GitHub data (optional)

• Check gh auth status to see if gh CLI is available
• If available, offer to pull: contribution calendar, repo list, language stats
• Reference: gh api graphql for contribution data, gh repo list for repos

4. Existing portfolio materials (optional)

• Any existing site content, project descriptions, or images to incorporate

Set Up the Repository

After collecting inputs, help the user create the portfolio repo:

1. Private source + public deploy (Recommended)

gh repo create {username}/portfolio --private --clone
cd portfolio
gh repo create {username}/{username}.github.io --public

• The {username}.github.io name is a GitHub Pages convention and must not be changed. The private source repo name (e.g., portfolio) can be anything.
• Two repos: portfolio (private) for source/reports, {username}.github.io (public) for deployed site
• Reports stay private, clean URL: https://{username}.github.io
• GitHub Actions builds in portfolio and pushes to {username}.github.io
• User controls which reports appear on the Raw Data page

2. Single public repo (Simple)

gh repo create {username}/{username}.github.io --public --clone
cd {username}.github.io

• One repo for everything — simplest setup
• All reports visible in git history
• Fine if user has no private content

Organize Reports

Copy reports into the repo:

reports/
├── {date}-{agent-name}/
│   ├── user-profile.md
│   ├── project-{name}.md
│   └── ...
└── aggregated/          (if Phase 2 was run)
    ├── aggregated-profile.md
    ├── references-multi-perspective.md
    ├── introduced-by-agents.md
    ├── faq.md
    └── blog-topics.md

If the user provides a resume, profile photo, or screenshots, save them to materials/. Any filename and format works (resume.pdf, 이력서.pdf, cv.md, etc.).

Portfolio Repo Structure

After setup, your repo should look like this:

portfolio/
├── reports/                    # Agent-reference reports (required)
│   ├── 2026-03-claude-code/
│   │   ├── user-profile.md
│   │   └── project-*.md
│   └── aggregated/             # Phase 2 outputs (if available)
│       ├── aggregated-profile.md
│       ├── references-multi-perspective.md
│       ├── faq.md
│       └── blog-topics.md
├── materials/                  # User-provided materials (optional)
│   ├── resume.pdf              # Resume/CV (any format and filename)
│   └── profile.jpg             # Profile photo, screenshots
└── README.md                   # Will be auto-generated

The skill will generate the src/, public/, and config files in the next steps. You only need to provide the data above.

Step 2: Analyze & Propose Concept

Read references/concept-generation-guide.md for the full concept generation workflow.

What You Do

1. Read all reports in the reports/ directory — user profiles, project reports, aggregated outputs
2. Identify dominant traits — work style signals, communication patterns, technical identity
3. Map to a design concept — each concept has a color palette, typography, and layout direction
4. Present the concept to the user for approval

Available Concepts

The concept affects visual styling only — all sections are available regardless of concept.

Always get user approval before proceeding. Show the proposed color palette, fonts, and section order. Offer to adjust or try a different concept.

Step 3: Scaffold Site

Read references/site-structure-guide.md for complete component specifications.

Initialize Astro

npm create astro@latest -- --template minimal --no-install
npm install
npm install tailwindcss @tailwindcss/vite

Warning: Do NOT use npx astro add tailwind — it installs @astrojs/tailwind (Tailwind 3 integration) which conflicts with Tailwind 4. Install @tailwindcss/vite directly instead.

Optional: Enhanced Design Quality

If the frontend-design skill is available in the current environment, invoke it during component implementation for production-grade styling and modern design patterns. This produces more polished, distinctive results — avoiding generic AI aesthetics.

• Available: Invoke frontend-design after concept approval, before generating components. Let it guide component styling, layout composition, and interaction design alongside the concept theme.
• Not available: The built-in concept themes and site-structure-guide provide a complete design system. The site will still look good.

This is an optional enhancement, not a dependency. The skill is never required to run agent-portfolio.

Generate Components

Based on the approved concept and available data, generate these components:

Apply Concept Theme

Configure concept-specific design tokens. Tailwind 4 supports both CSS-based configuration (using @theme in your CSS) and the traditional tailwind.config.mjs. Either approach works — see site-structure-guide.md for details. Import Google Fonts in Layout.astro.

Process Report Data

Create src/lib/parse-reports.ts to transform raw markdown reports into structured data for components. See site-structure-guide.md for the parsing functions specification.

Step 4: Customize & Review

Preview Locally

npm run dev

Open http://localhost:4321 and walk through the site with the user.

Review Checklist

Go through each section with the user:

• Hero — Does the summary line capture who they are?
• Agent Reviews — Are the persona voices distinct and accurate?
• Projects — Are the right projects highlighted? Any to add/remove?
• Work Style — Does the visualization feel accurate?
• Overall feel — Does the concept match their identity?

Pre-deployment Privacy Check

Before deploying, review the site content for sensitive information. The reports were written for the user's reference — not all content is safe to publish. Walk through this checklist with the user:

• PII — No phone numbers, email addresses, or home addresses on the site (check resume-derived content especially)
• Private repo details — Reports for private/employer projects should match their assigned privacy level (Full/Private/Stats-only). Check that private projects don't leak identifying details (specific domain names, internal URLs, unique tech combinations)
• Business metrics — No exact user counts, MAU, revenue, or cost figures from private projects (use qualitative descriptions)
• Security details — No RLS function names, auth implementation patterns, SECURITY DEFINER details, or error filtering rules from production services
• Employer internal systems — No internal tool names, infrastructure architecture, or vendor integrations that could violate NDA or expose attack surface
• Materials — Resume PDF is NOT served as a static file (extract content only, don't copy the file to public/)

Adjust

Based on feedback:

• Content changes — update report excerpts, rewrite summaries
• Design changes — adjust colors, fonts, section order in Tailwind config
• Section changes — add/remove optional sections
• Concept change — if the whole direction feels wrong, go back to Step 2

Iterate until the user approves the site.

Step 5: Deploy

Read the deployment guide matching the user's chosen option:

• Option 1 (private source + public deploy): references/deployment-private.md
• Option 2 (single public repo): references/deployment-public.md

Quick Deploy

For Option 1 (private source + public deploy), set up the deploy key first (see references/deployment-private.md), then:

git add -A
git commit -m "Initial portfolio site — Introduced by My Agents"
git push -u origin main

GitHub Actions builds in portfolio and pushes to {username}.github.io. The site will be live at https://{username}.github.io.

Verify

• Check the Actions tab in the source repo for build status
• Visit https://{username}.github.io once deployment completes
• Test on mobile (responsive check)

GitHub Profile README (Optional)

If the user wants, generate a minimal GitHub Profile README that links to the portfolio. See the deployment guides for the template.

Updating the Site

When you have new reports (new projects, new machine, new agent):

1. Run agent-reference skill again → new report files
2. Copy new reports to reports/ in the portfolio repo
3. Run this skill again or just npm run build to regenerate
4. Push → GitHub Actions rebuilds automatically

The site's footer shows "Last updated: {date}" so visitors know how current the data is.

Important Principles

The site tells the user's story through agent eyes. Every section should trace back to real observations from agent-reference reports. Don't pad with generic content.

Concept serves content. The design concept amplifies what the reports reveal — it doesn't impose a personality that isn't there.

User controls everything. Which reports to include, which sections to show, what to make public. The agent proposes; the user decides.