skillgrowth

A self-hosted tool for your career and skill growth that grows together with you, instead of asking you to fill out a static profile once.

Who it's for: working professionals building their career over the long term — but it's especially useful if you're actively job hunting or considering a change, since your skills, activity, and history are already organized into a ready-to-use resume the moment you need one, instead of having to reconstruct them from memory under time pressure.

A note on terminology: every "skill" in this app is one of your own career/professional skills (e.g. "Python", "public speaking", "project management") — not an AI "skill" in the sense of a Claude Code Skill, an LLM plugin, or an agent capability. skillgrowth tracks skills that belong to you as a person; the LLM it optionally talks to is just a tool it uses to extract and match them, not a source of skills itself.

Why

Your career history keeps ending up on someone else's platform. During a job search, it's a recruiter's site. During a performance review, it's your employer's internal HR system. Either way, that data was never really yours — and when you leave, it stays behind. What you're actually left holding, if anything, is a handful of scattered Word and Excel files you happened to save yourself.

skillgrowth exists to fix that: your career record — skills, evidence, history, resume — lives on your own server, under your own control, built up on your own initiative rather than only when a job search or an annual review forces you to think about it. The goal is a skill set that's genuinely yours, portable across employers, not something a company's database happens to be holding onto this year.

Every piece of activity you feed it — a quick update, a certification photo, an education/employment/project record, a reading/talk/certification entry — is kept as an append-only activity log. An LLM extracts skills from that activity and matches them against skills you already have, so your skill picture and a ready-to-use resume can be derived from the log at any time — the resume itself is generated from a fixed template, no LLM required. Skills can also be added directly (one at a time, or in bulk via CSV) when you don't have free-text activity to extract from, or drafted in bulk from an existing resume (reviewed and confirmed before anything is saved) if you'd rather not type years of history in by hand.

flowchart LR
  U1[Quick update] --> E[(Activity log)]
  U2[Certification image] --> E
  U3[Education / Employment / Project] --> E
  U4[Reading / talk / certification entry] --> E
  U5[Manual add / CSV import] --> S[(Skill)]
  U6[Resume import - reviewed draft] --> E
  U6 --> S
  E --> L[LLM extraction + matching]
  L --> S
  S --> V[Current skill view]
  S --> X[Resume export - no LLM]
  S --> G[AI Career Support: career consult / job gap check / growth guidance]

Loading

See docs/ARCHITECTURE.md for the full data model and request flow.

Features

Full feature list: docs/FEATURES.md (日本語).

• Dashboard — career path goals (this year / 5 years / 10 years, optional, editable with a full — paginated — history of past edits), a monthly bar chart of skill-improving actions, a category breakdown chart, and a quick update box.
• Vision — one free-form text box for a rough sketch of the kind of career you're aiming for, with no time horizon or structure — a looser complement to the Dashboard's three specific goals.
• Self Feedback — periodic self-review on any date you choose: what you did, your reflection on it, and what to carry forward, side by side. Fully editable and deletable, unlike the append-only activity log elsewhere in the app.
• Profile — education, employment, and projects (standalone or linked to an employer), each addable, editable in place, and deletable. Free-text descriptions also feed skill extraction.
• Portfolio — deliverables and work samples: title, description (not fed into skill extraction), any number of links, and any number of uploaded files (PDF, spreadsheets, photos, up to 10MB each). Can optionally link to a standalone project, never one tied to an employer.
• Skills — the current skill picture derived from the activity log; supports adding a skill directly by name, editing a skill's name/category in place, or importing a batch from a name,category CSV file.
• Resume Import — upload an existing resume (.docx or .pdf) and let the LLM draft Education/Employment/Project/Certification/Self-PR entries from it. Nothing is saved until you review and confirm each item; skills found in the resume get linked to the project(s) they came from (so they show up in Skills and the Skill Network) rather than registered as a disconnected list.
• Activity — add a reading/talk/certification entry (with optional certificate image upload for OCR extraction) and browse a chronological feed of everything that's been added, in one page.
• Certifications — a dedicated page listing every certification with its acquisition date and an optional expiry date (flagged once past), separate from Activity's own "Certifications only" filter — this is also the only place a certification's structured fields can be edited after the fact.
• Skill Network — a force-directed graph showing how skills connect to the certifications, education, employment, projects, and portfolio pieces that back them. Deliberately excludes day-to-day activity entries (reading, talks, quick updates), so the graph stays readable no matter how long you've used the app. Label wrap width, node spacing, and edge length are adjustable and remembered per-browser.
• Resume — a Self PR field (kept as paginated history, most recent used) plus a resume generator that fills a fixed Markdown template from your current data — no LLM involved, rendered and downloadable, with past generations kept as browsable snapshots, any of which can be edited directly as Markdown afterward. Alongside that, upload your own .docx template with tags like {{p self_pr }} and generate a filled copy of it — also no LLM involved, with per-section bullet-list/table layout chosen from the UI.
• AI Career Support — the only features that call an LLM on demand: Career Consult (a saved, multi-turn chat with an AI career consultant, grounded in your actual skills/activity/history/goals on every message), growth guidance toward each Dashboard goal, and a job-posting gap check (paste a job description to see which of its requirements you already meet).
• Settings — language, theme (light/dark/system) and an accent color, the LLM connection (base URL / API key / model, with a test button), a skill extraction toggle (off by default — turn it on to have Quick update/Activity/Profile text automatically run through the LLM; off, those saves are instant with no LLM call), a full data backup download, restoring from a backup file, and an About panel with the app version and license.
• A Contribute button lives in the sidebar on every page (above the copyright line) — skillgrowth is MIT-licensed and issues/PRs are genuinely welcome.

Screenshots

Quick update → skill extraction → the extracted skills showing up on the Skills page, the core evidence-to-skill loop the "Why" section above describes:

The rest of the screens below are shown with the built-in sample data (Settings → "Try it with sample data"), not a real account.

Dashboard	Skills

Activity	Resume

Resume Import	Skill Network

Design choices

• Single user, self-hosted. No auth, no multi-tenancy. Run your own instance the way you'd self-host a personal finance tool. If you expose it to the internet rather than keeping it on a local network or VPN, put something in front of it — see "Optional access gate" below, or your reverse proxy's own auth.
• Free-text skills. No fixed taxonomy. Skills are stored as the natural language the LLM extracts from your activity (or that you type directly), not normalized IDs.
• LLM-optional, not LLM-required. Extracting skills from free text and the AI Career Support features are the only things that call an LLM. Adding/editing skills by hand, career path goals, Vision, Self PR, resume generation, and backup/restore all work with none configured. Extraction itself is also off by default even once an LLM is configured — a separate Settings toggle you turn on explicitly, so a slow local model never blocks a save unless you've asked for it to.
• Pluggable LLM. When a feature does call one, it talks to any OpenAI-compatible chat completions endpoint — a cloud API or a local Ollama server (http://localhost:11434/v1). Configured from the Settings page.
• Activity, not self-assessment. Skills mostly come from things you already have (certificates, project notes, a CSV export from wherever you already track this) rather than from filling out a skill matrix — though you can also add a skill by hand.

Quick start

docker compose up --build

Or, without cloning the repo, run the prebuilt image directly:

docker run -d -p 8000:8000 -v $(pwd)/data:/app/data ghcr.io/tkm112345/skillgrowth:latest

Open http://localhost:8000, then set your LLM connection under Settings. Want to see what a populated app looks like first? Settings → "Try it with sample data" loads a small fictional career history with one click.

Example LLM configurations

Any OpenAI-compatible Chat Completions endpoint works. A few confirmed to work as of this writing:

Provider	Base URL	API key	Notes
OpenAI	https://api.openai.com/v1	an OpenAI API key	the default
Ollama (local)	http://localhost:11434/v1	anything non-empty	run a model locally, no external calls
Claude (Anthropic)	https://api.anthropic.com/v1/	an Anthropic API key	Anthropic's docs describe this OpenAI-compatible layer as being for quick evaluation, not a long-term production integration — some features (e.g. prompt caching) aren't available through it
Gemini (Google)	https://generativelanguage.googleapis.com/v1beta/openai/	a Gemini API key from Google AI Studio	documented as beta by Google

None of these let you authenticate with a consumer subscription login (e.g. a Claude Pro/Max or ChatGPT Plus account) instead of a billed API key — neither Anthropic nor OpenAI allow that for third-party applications, so a metered API key is the only option regardless of provider.

Optional access gate

skillgrowth still has no user accounts — this is a single shared password, not a login system. When both SKILLGROWTH_BASIC_AUTH_USER and SKILLGROWTH_BASIC_AUTH_PASS environment variables are set, every request (except /api/health, so the Docker HEALTHCHECK keeps working) requires that username/password over HTTP Basic Auth. Unset (the default), the app is fully unauthenticated, unchanged from before:

docker run -d -p 8000:8000 -v $(pwd)/data:/app/data \
  -e SKILLGROWTH_BASIC_AUTH_USER=youruser \
  -e SKILLGROWTH_BASIC_AUTH_PASS=yourpassword \
  ghcr.io/tkm112345/skillgrowth:latest

This is a coarse, single-password gate, not a real auth system — fine for "keep casual visitors and bots out of an internet-facing instance," not a substitute for a reverse proxy's own auth (or a VPN/local network) if you need anything stronger.

Usage

A typical first session looks like this:

1. Settings → LLM connection. Point it at OpenAI, another OpenAI-compatible provider, or a local Ollama server, then "Test connection." Skip this if you only want the LLM-free parts of the app (Resume, backup/restore, browsing) for now — you can come back to it later. Skill extraction itself is a separate toggle on the same page, off by default — turn it on if you want step 3 below to actually extract skills from what you write.
2. Settings → Try it with sample data, if you want to see a populated app before typing anything yourself. A matching "Reset sample data" button removes exactly what this added, whenever you're ready to start for real.
3. Dashboard → Quick update. Write a sentence or two about something you've been working on. With skill extraction turned on (step 1), this is the fastest way to see the evidence → extraction → skill loop in action: submit it, and any skills the LLM recognized show up immediately. With it off, the same submit just saves your update to the activity log instantly, with no skills extracted.
4. Activity, Profile, and Skills are the other ways to feed the same loop: log a certification or a book you read on Activity, fill in education/employment/projects on Profile (their free-text fields feed extraction too), or add a skill directly on Skills if you already know you have it and don't need evidence for it.
5. Resume Import, if you already have a resume: upload it (.docx or .pdf) and let the LLM draft Education/Employment/Project/Certification/ Self-PR entries from it instead of typing years of history in by hand — nothing is saved until you review and confirm each item.
6. Dashboard → Career path goals and Vision, whenever you want to write down where you're headed — goals are time-boxed (this year / 5 years / 10 years) and keep a full edit history; Vision is one looser, unstructured paragraph with no history, for whatever doesn't fit into "by when."
7. Resume, once you have some Profile/Activity data: write a Self PR pitch, then "Generate resume from current data." No LLM call — it's your data poured into a fixed template — so this works even before step 1. Every generation is kept, so you can always go back to an earlier version.
8. AI Career Support, only if you configured an LLM in step 1: start a Career Consult conversation, paste a job posting for a gap check against your current skills, or ask for growth guidance toward the goals you wrote in step 6.
9. Settings → Data backup, occasionally: downloads everything (except the LLM connection settings) as one JSON file. Restoring it later — on this instance or a fresh one — is purely additive, so it's safe to import into an install that already has data.

Local development

See docs/CONTRIBUTING.md for reporting bugs, opening pull requests, and running the backend and frontend separately with hot reload. See docs/ROADMAP.md for what's left before this project calls itself v1.0.0.

Known limitations

• Certification ingestion accepts images or a .pdf (only its first page is read — a PDF is converted to one image, not treated as a multi-page document).
• Resume Import reads .docx or .pdf (text only — a scanned/image-only PDF has no extractable text and is rejected); no parsing of arbitrary documents beyond that one fixed flow.
• The "current skills" view is read from the database directly, but the match/merge step that keeps it deduplicated runs at evidence-ingestion time via an LLM call — quality depends on the configured model.
• The Markdown resume generator follows a fixed set of sections and isn't customizable per-export; edit the downloaded Markdown by hand for anything beyond that, or use the Word template export for a custom layout instead.

Changelog

See CHANGELOG.md.

License

MIT — see LICENSE.