AgentFolio

An open-source agentic portfolio engine. Fork it, drop in your resume, and deploy a portfolio site that adapts to each visitor's context.

Live Demo · Dashboard Demo

How It Works

AgentFolio renders a resume adapted to the target company and role. A three-stage pipeline (/fit → /extract-directives → /structurize) tailors your resume, lets you edit the result in markdown, and converts it to JSON Resume for rendering and PDF export. Each URL slug maps to a tailored adaptation.

/                    → default resume
/company-slug        → company-specific adaptation
/dashboard           → owner dashboard (not public)
/unknown             → 404 page

Quick Start

1. Fork this repo
2. Enable GitHub Pages: go to Settings → Pages → Source → select GitHub Actions
3. Replace data/input/resume.md with your resume (any format — paste from LinkedIn, PDF text, or write markdown)
4. Add target positions in data/input/jd/ — one .md file per role, filename becomes the URL slug (e.g., data/input/jd/google.md → yoursite.com/google)
5. Set secrets on your fork:
   • CLAUDE_CODE_OAUTH_TOKEN — uses your Claude Pro/Team subscription instead of usage-based API billing
   • WORKFLOW_PAT — a fine-grained or classic PAT with actions: write (or classic workflow) scope, so bot workflows can chain (fit → structurize → pdf → deploy). Without it, each stage's "Trigger …" step 403s and the chain stops after fit.
6. Push — GitHub Actions generates adapted resumes, PDFs, and deploys to GitHub Pages

No local runtime needed. No JSON to write. Just markdown and push.

Personalization

All personal data lives in data/input/. Replace these files with your own:

File	Purpose
data/input/resume.md	Your resume in any text format
data/input/jd/*.md	Target positions (one .md per role, filename = URL slug)
data/input/directives.md	Global resume preferences (e.g., "prefer platform engineer over backend engineer")
data/fitted/*.md	Tailored markdown — auto-generated by /fit, human-editable
data/adapted/*.json	Final JSON Resume — auto-generated by /structurize, don't edit
data/adapted/*.pdf	Auto-generated PDFs — don't edit

Framework code in web/ and .github/ is generic — you shouldn't need to modify it.

Environment Variables

Set these in web/.env.local for development, or as GitHub Actions secrets/env for production:

Variable	Purpose
VITE_GITHUB_REPO	your-username/your-repo. Auto-set in deploy workflow via ${{ github.repository }}.
VITE_BASE_PATH	URL base path. Auto-detected from actions/configure-pages: / for user pages or custom domains, /<repo-name>/ for project pages.
CLAUDE_CODE_OAUTH_TOKEN	OAuth token for Claude Code in the adapt workflow. Uses your Claude Pro/Team subscription instead of usage-based API billing.
WORKFLOW_PAT	PAT with actions: write (fine-grained) or workflow (classic) scope. Used by bot workflows to dispatch downstream stages (fit → structurize, extract → structurize, structurize → deploy, pdf → deploy). The default GITHUB_TOKEN lacks this permission and 403s on cross-workflow dispatch.
VITE_CHAT_PROXY_URL	URL of the deployed chat Worker. If unset, the chat panel renders an offline card with contact links.

Enable chat (optional)

Adds a floating "Chat with me" widget on every public slug, backed by Claude. Visitors ask questions about you and get first-person streaming answers grounded in that slug's fitted resume, directives, and JD.

Skipping these steps leaves your site identical to the baseline.

1. cd proxy && npm install && npm i -g wrangler && wrangler login
2. Create a KV namespace: wrangler kv namespace create RATE_LIMIT. Paste the returned id into proxy/wrangler.toml.
3. Set NAME under [vars] in proxy/wrangler.toml to your display name.
4. Set secrets: wrangler secret put ANTHROPIC_API_KEY, ALLOWED_ORIGIN, PAGES_ORIGIN, IP_HASH_SALT.
5. wrangler deploy. Copy the *.workers.dev URL.
6. Set VITE_CHAT_PROXY_URL on your Pages env (repo Settings → Secrets and variables → Actions → Variables) and redeploy.

See proxy/README.md for full configuration reference.

Pipeline

data/input/resume.md + jd/anthropic-fde-nyc.md + directives.md
        │
        ▼  /fit
data/fitted/anthropic-fde-nyc.md          ← tailored markdown (human-editable)
        │
        ▼  /extract-directives  (only if you edited the fitted file)
data/input/directives.md       ← preferences auto-extracted from your edits
        │
        ▼  /structurize
data/adapted/anthropic-fde-nyc.json       ← JSON Resume for rendering
data/adapted/anthropic-fde-nyc.pdf        ← PDF export

/fit reads your resume, a target JD, and your directives. Outputs tailored markdown you can review and edit.

/extract-directives diffs your edits against the last generated version and appends the intentions to directives.md so they persist across re-runs.

/structurize converts the fitted markdown into a valid JSON Resume document for the web theme and PDF export.

Growing your base

/grow-resume expands data/input/resume.md through a structured conversation so /fit has richer material to work with. Section-by-section walkthrough, gap audit against directives, integrity checks against fabrication. Run it once to set a career goal (persisted in data/input/career-goal.md) and any time afterward to iterate a specific section. Orthogonal to the pipeline — grow the base first, then /fit uses the expanded version.

Features

Owner dashboard for reviewing fitted resumes, diffs, PDFs, and JDs

• Three-stage pipeline — separate tailoring from schema conversion, with human editing in between
• Directive learning — your edits are automatically extracted as reusable preferences
• Preset directives — research-backed resume optimization rules out of the box
• Owner dashboard — preview, word-level diff, inline PDF, and JD tabs for each fitted resume
• Fit summaries — auto-generated change descriptions for each adaptation
• PDF export — auto-generated PDFs alongside each adaptation, with a download button on the site
• Zero-runtime quickstart — fork, add markdown files, push, deployed. No local tools needed.
• JSON Resume theme — renders all 12 JSON Resume sections using the developer-mono theme
• Live GitHub activity — daily cron pulls contribution heatmap, top languages, and recent repos into a dynamic section below the resume. Independent of the resume pipeline.

Architecture

web/              React SPA (Vite + TypeScript + styled-components)
.claude/skills/   Claude Code skills (grow-resume, fit, extract-directives, structurize)
data/input/       Your personal data — resume, JDs, directives
data/fitted/      Tailored markdown (auto-generated, human-editable)
data/adapted/     Final JSON Resume + PDFs (auto-generated)
.github/          GitHub Actions workflows (fit, extract, structurize, deploy, pdf)

Credits

AgentFolio stands on the shoulders of open source. Thanks to:

Resume stack

• JSON Resume — the open resume schema that every adaptation conforms to
• jsonresume-theme-developer-mono by Thomas Davis — web/src/components/ResumeTheme.tsx is adapted from this theme
• jsonresume-theme-onepage by Ainsley Chong — PDF theme used by the pdf workflow
• resumed — CLI that drives PDF export

Framework

• React, Vite, TypeScript, styled-components, react-markdown, diff
• Vitest, Playwright — test runners
• Puppeteer — PDF rendering inside the workflow
• Cloudflare Workers — chat proxy runtime

Agentic engine

• Claude Code by Anthropic — powers the /fit, /extract-directives, and /structurize skills

License

MIT