Skip to content

amancodingrepo/skillgod

Repository files navigation

SkillGod

Claude knows how to code. SkillGod knows how you code.

SkillGod is a runtime that sits between you and your AI coding tool. Before every response, it injects the right skills, your project memory, and routes complex tasks to specialist agents — automatically.

No prompt engineering. No copy-pasting context. It just works.


The problem

Claude is brilliant out of the box. But it doesn't know:

  • Your team's debugging methodology
  • Which deployment pattern your project uses
  • What you decided last Tuesday about state management
  • That the last time it suggested Redux, you reworked it three times

Every session starts cold. SkillGod fixes that.


What changes

Without SkillGod With SkillGod
Context per prompt 0 3 scored skills + 32 instincts
Memory across sessions Cold start every time SQLite per project
Agent routing One model, one task 10 specialist agents, each with own skills
Vault 1,944 skills
Learns from your feedback No Yes (accept/rework signals)
Prompt injection protection No Yes (security scan on every input)

How it works

Your prompt
    │
    ▼
┌──────────────────────────────────────────────┐
│  Security scan — block injections first      │
└──────────────────────────────────────────────┘
    │
    ▼
┌──────────────────────────────────────────────┐
│  Pillar 1 — Skills                           │
│  Score 1,944 skills against your task        │
│  Inject top 3 + all 32 instincts             │
│  (instincts fire every single prompt)        │
└──────────────────────────────────────────────┘
    │
    ▼
┌──────────────────────────────────────────────┐
│  Pillar 2 — Memory                           │
│  Load decisions, patterns, errors            │
│  for this specific project                   │
│  Record accept/rework signals after          │
└──────────────────────────────────────────────┘
    │
    ▼
┌──────────────────────────────────────────────┐
│  Pillar 3 — Agents (complex tasks only)      │
│  Decompose → spawn specialists               │
│  Frontend agent gets UI skills               │
│  Backend agent gets API skills               │
│  Nobody else does this                       │
└──────────────────────────────────────────────┘
    │
    ▼
Claude (or any AI tool)

Install

# macOS (Apple Silicon)
curl -L https://skillgod.dev/dist/sg-mac-arm64 -o /usr/local/bin/sg && chmod +x /usr/local/bin/sg

# macOS (Intel)
curl -L https://skillgod.dev/dist/sg-mac-intel -o /usr/local/bin/sg && chmod +x /usr/local/bin/sg

# Linux
curl -L https://skillgod.dev/dist/sg-linux -o /usr/local/bin/sg && chmod +x /usr/local/bin/sg

# Windows (PowerShell)
irm https://skillgod.dev/install.ps1 | iex

# From source
git clone https://github.com/skillgod/skillgod
cd skillgod
pip install -r requirements.txt
python engine/ingest.py
sg init

Then restart Claude Code. Skills inject automatically via MCP — no other config needed.


Quick start

sg init                          # detects IDE, writes .mcp.json, rebuilds index
sg find "debug python traceback" # test skill discovery
sg stats                         # vault counts and health
sg scan "your prompt here"       # test security scanner

The vault

1,944 skills across 11 categories. Every skill has:

  • Trigger words for accurate scoring
  • A description that starts with "Use when..." (not a summary — this is what makes discovery work)
  • Confidence score (0.65–0.95)
  • Source attribution
Category Skills What's in it
coding 572 debugging, TDD, Python, Go, Rust, review patterns
agents 461 frontend, backend, devops, security, docs specialists
design 250 UI/UX pro, brand, design systems, Figma patterns
writing 168 technical docs, README, blog, comms
devops 157 Docker, Kubernetes, Terraform, CI/CD, cloud deploy
instincts 32 always-on rules — verify before done, security first
research 100 deep research, market analysis, summarisation
security 96 OWASP, audit, injection detection, threat modelling
api 79 REST, GraphQL, API design, SDK patterns
react 29 hooks, composition, React Native, Expo
meta auto-learned skills pending review

Commands

sg init              Set up SkillGod — detects IDE, writes .mcp.json
sg find <task>       Score vault against a task, show top matches
sg stats             Vault health, category counts, index status
sg build             Interactive skill builder — guided prompts
sg learn             Save any output as a new skill (goes to meta/)
sg sync              Rebuild local index (free) or decrypt full vault (pro: sg sync --key YOUR-KEY)
sg scan <text>       Scan for prompt injection threats
sg signals           Accept/rework rates per skill — see what's working
sg promote           Review auto-learned skills, approve for vault promotion
sg run <tool>        Run any AI tool with SKILLGOD_ROOT injected

MCP tools (what Claude Code calls directly)

Once sg init runs, these tools are available inside every Claude Code session:

find_skills(task)         → top scored skills for this task
inject_context(task)      → full augmented prompt string
save_memory(summary)      → persist a decision to project memory
get_memory(project)       → retrieve recent project context
learn_skill(task, output) → auto-generate skill from session output
stocktake()               → vault health report
spawn_agents(task)        → decompose and route to specialist agents
security_scan(text)       → threat detection
vault_stats()             → live vault counts

Pricing

Feature Free Early Adopter Pro
Price Free forever $7/mo locked forever $10/mo
Who Everyone First 200 users only User 201+
Skills 30 starter skills 885+ (full vault) 885+ (full vault)
Instincts All 32 All 32 All 32
Memory Full SQLite layer Full SQLite layer Full SQLite layer
Agent layer Basic Full multi-agent Full multi-agent
Monthly vault updates No Yes Yes
Skill enhancer + signal analytics No Yes Yes
sg sync --key No Yes Yes

The free tier is not crippled — 30 skills, full memory, and the basic agent layer is enough to genuinely feel the difference.

Early Adopter pricing is locked forever. First 200 users pay $7/mo for life, no matter what the price does later.

Referral program:

  • Get a referral link from any Pro user → sign up at $7/mo locked forever, even after the 200-user window closes
  • Refer a friend who converts to paid → you get 1 free month ($10 value) automatically

License via LemonSqueezy. Offline grace period: 30 days — your workflow never breaks.


Why the vault is the moat

Anyone can build a skill injector. The vault is 1,944 curated, scored, and formatted skills that have been normalised from 12 source repos into a single consistent format. Every description starts with "Use when..." — not a summary — which is the detail that makes semantic scoring work.

The free runtime is open source. The vault is the product.


Built on the shoulders of

SkillGod assembles the best patterns from 12 open-source repos:

Repo What we took
obra/superpowers SKILL.md format, "Use when" description standard
anthropics/skills 17 production skills, marketplace format
thedotmack/claude-mem SQLite memory schema, 5-hook lifecycle
affaan-m/everything-claude-code Instincts layer, confidence scoring
msitarzewski/agency-agents 179 specialist agent profiles
snarktank/ralph Autonomous loop, circuit breaker pattern
nextlevelbuilder/ui-ux-pro-max-skill Deep design intelligence — 161 palettes, 57 font pairings
vercel-labs/skills Deployment patterns
vercel-labs/agent-skills Agent-specific skill patterns
hesreallyhim/awesome-claude-code Hook catalog, slash-command patterns
yusufkaraaslan/Skill_Seekers Ingestion pipeline design
VoltAgent/awesome-agent-skills Official team skills from Stripe, Cloudflare, Netlify, HuggingFace

Architecture

skillgod/
├── engine/
│   ├── ingest.py        ← sources → vault normaliser
│   ├── memory.py        ← SQLite memory layer
│   ├── skills.py        ← scoring + injection engine
│   ├── agents.py        ← spawning + per-agent skill routing
│   ├── runtime.py       ← combines all three pillars
│   ├── mcp_server.py    ← FastMCP server on :3333
│   ├── security.py      ← injection detection
│   ├── encryption.py    ← AES-256-GCM vault encryption
│   ├── signals.py       ← accept/rework tracking
│   └── variants.py      ← promotion queue
├── hooks/
│   ├── session_start.py ← loads instincts + memory
│   ├── pre_tool.py      ← security scan + skill injection
│   └── post_tool.py     ← captures memory + learns skills
├── vault/               ← 1,944 normalised skills (open)
├── cli/                 ← Go binary: sg
└── deploy/              ← Railway API stub (future backend)

MIT License. Vault content retains original licenses from source repos.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages