---
title: SkillWiki
emoji: πΈοΈ
colorFrom: blue
colorTo: indigo
sdk: docker
pinned: true
license: mit
app_port: 7860
---
# SkillWiki
**A Living Knowledge Infrastructure for Agent Skills**
Just as Wikipedia organizes human knowledge and GitHub governs software, **SkillWiki** provides the missing infrastructure for agent skills. It transforms heterogeneous knowledge sources β trajectories, documents, API specs, and execution experience β into reusable skill assets that are grounded in their originating evidence, version-controlled, and continuously evolved. SkillWiki covers the complete skill lifecycle: from knowledge ingestion and skill production, through provenance-aware graph exploration and human governance, to execution-driven health monitoring and automated evolution. The result is a shared infrastructure where knowledge, skills, and execution experience co-evolve together.
[](LICENSE)
[](https://www.python.org/)
[](https://nodejs.org/)
[](https://github.com/Huangdingcheng/SkillWiki)
[πΊ **Demo Video**](https://www.youtube.com/watch?v=TODO) Β·
[π **Paper**](https://arxiv.org/abs/2606.16523)
---
SkillWiki system overview β from multi-source ingestion to governed skill evolution.
---
## Key Features
| | Feature | Description |
|:---:|---|---|
| π₯ | **Multi-source ingestion** | Trajectories, documents, API specs, scripts, and existing skill JSON/JSONL files |
| π | **Governed lifecycle** | Eight-stage state machine (S0βS7) with automated verification and full audit trail |
| πΈοΈ | **Provenance knowledge graph** | Typed edges β `depends_on`, `composes_with`, `evolved_from`, `replaces`, and more |
| π©Ί | **Health & evolution engine** | Continuous monitoring detects degraded skills and queues repair proposals automatically |
| π | **Bilingual UI** | English (default) and Chinese β toggle via the language button in the header |
| β¨οΈ | **CLI-first** | Every operation available as a `skillwiki` command for scripting and agent integration |
---
## Lifecycle
```
S0 Raw βββΊ S1 Candidate βββΊ S2 Draft βββΊ S3 Verified βββΊ S4 Released
β
S7 Archived βββ S6 Deprecated βββ S5 Degraded
```
| State | Name | Meaning |
|:---:|---|---|
| **S0** | Raw | Ingested source material, not yet extracted |
| **S1** | Candidate | Extracted skill candidate awaiting review |
| **S2** | Draft | Formalized schema, pending verification |
| **S3** | Verified | Passed automated postcondition checks |
| **S4** | Released | Approved for agent use in production |
| **S5** | Degraded | Success rate fell below health threshold |
| **S6** | Deprecated | Replaced or retired |
| **S7** | Archived | Read-only historical record |
---
## Quick Start
### Requirements
- Python 3.10+
- Node.js 18+ *(frontend only)*
### Install
```bash
git clone https://github.com/Huangdingcheng/SkillWiki.git
cd SkillWiki/skillwiki
python -m venv venv
source venv/bin/activate # Linux / Mac
venv\Scripts\activate # Windows
pip install -r requirements.txt
pip install -e .
```
### Start the backend
```bash
skillwiki serve --port 8001 --api-key YOUR_LLM_API_KEY
```
### Start the frontend *(optional)*
```bash
cd ../skillwiki-frontend
npm install
npm run dev # β http://localhost:3000
```
### One-click launcher (Windows)
Double-click **`START_SKILLWIKI_DEMO.bat`** in the repository root. On first run you will be prompted for your LLM endpoint and API key (saved to `skillwiki-launcher\config.local.ps1`, which is Git-ignored β never commit it).
```
Backend: http://127.0.0.1:8001
Frontend: http://127.0.0.1:5174/wiki
```
Restore the public demo fixtures after a restart:
```
RESTORE_SKILLWIKI_DEMO_STATE.bat
```
---
## Demo Walkthrough
| Step | Page | What to explore |
|:---:|---|---|
| 1 | [`/wiki`](http://127.0.0.1:5174/wiki) | Skill library β browse, filter by state / tag, inspect schemas |
| 2 | [`/ingest`](http://127.0.0.1:5174/ingest) | Paste an API doc or upload a file; watch candidates extracted live |
| 3 | [`/graph`](http://127.0.0.1:5174/graph) | Provenance knowledge graph β Nebula, Readable, and Debug view presets |
| 4 | [`/harness`](http://127.0.0.1:5174/harness) | Execute-verify loop; observe automated repair attempts |
| 5 | [`/evaluation`](http://127.0.0.1:5174/evaluation) | SkillsBench P0 sparse-subset analysis |
| 6 | [`/versions`](http://127.0.0.1:5174/versions) | Business-readable diffs and re-verification after changes |
---
## CLI Reference
Install once after cloning:
```bash
cd skillwiki && pip install -e .
```
All commands accept `--api-url ` (default `http://127.0.0.1:8001`).
### Server
```bash
skillwiki serve [--host HOST] [--port PORT] [--backend memory|sqlite|postgres]
```
### Knowledge Ingestion
```bash
# source_type: trajectory | document | api_doc | script | past_skills
skillwiki ingest run [--create]
skillwiki ingest run api_doc ./openai_spec.md --create
skillwiki ingest run trajectory "open browser -> search -> copy link"
skillwiki ingest run past_skills ./skills.json --max-candidates 20
skillwiki ingest status
```
### Skill Lifecycle
```bash
skillwiki skill list [--state S3] [--tag nlp] [--limit 20]
skillwiki skill get [--full]
skillwiki skill status
skillwiki skill exec --input '{"key": "value"}'
skillwiki audit
skillwiki verify [--harness mock|claude_code|codex] [--max-retries 3] [--watch]
skillwiki promote
```
### Health & Evolution
```bash
skillwiki health [--json] # system-wide overview
skillwiki health [--json] # success rate, issues, open proposals
skillwiki repair # generate maintenance candidate
skillwiki evolve [--json] # run one full evolution cycle
```
### Maintenance Proposals
```bash
skillwiki proposal list [--status pending|accepted|rejected] [--json]
skillwiki proposal accept
skillwiki proposal reject
```
### Knowledge Graph
```bash
skillwiki graph neighbors [--depth 1]
skillwiki graph show [--view skill_only|provenance|version_impact] [--depth 2]
skillwiki graph deps
skillwiki graph export [-o output.json] [--view provenance] [--depth 2]
```
### Natural Language Execution
```bash
skillwiki run "summarize the attached PDF and extract action items" [--verbose]
```
---
## What SkillWiki Provides
- Knowledge-to-skill transformation from heterogeneous sources.
- Evidence-grounded skill assets with explicit provenance.
- Multi-granularity skill organization and dependency management.
- Provenance-aware exploration and visualization.
- Versioned skill governance with interpretable evolution.
- Execution-driven verification and continuous improvement.
- End-to-end demonstrations of reusable skill ecosystems.
---
## Environment Variables
| Variable | Default | Description |
|---|---|---|
| `LLM_API_KEY` | β | LLM API key *(required)* |
| `LLM_API_URL` | `https://api.deepseek.com` | LLM base URL |
| `LLM_MODEL` | `deepseek-v4-flash` | Model name |
| `SKILLOS_API_TARGET` | `http://127.0.0.1:8001` | Frontend proxy target |
---
## License
[MIT](LICENSE) Β© 2026 SkillWiki Authors