SkillGod runs three processes silently before every prompt:
1. **Skills** — Your task is scored against 1,944 curated skills in under 15ms. The top 3 matching skills inject automatically. You never choose them.
2. **Memory** — Every decision, pattern, and fix is captured to a local SQLite database. Tagged to the git branch and commit it happened on. Next session it is already there.
3. **Agents** — Complex tasks decompose automatically. Frontend agent gets UI and React skills. Backend agent gets API and validation skills. Both run simultaneously.
CLI Commands
sg runStart SkillGod runtime (auto-starts on IDE open)
sg find <task>Score and list skills that match a task description
sg learnManually trigger skill learning from recent session
sg sync --key KEYSync Pro vault using your license key
sg buildRebuild the local skill index from vault files
sg memoryView captured memory for current project
sg signals enableOpt in to anonymous quality signal collection
sg signals disableOpt out of signal collection
sg statusShow runtime status and loaded skill count
Memory Layer
Memory is stored in a local SQLite database at `~/.skillgod/memory.db`. It is never sent to any server.
**Memory kinds:**
- `decision` — architectural choices, library picks, approach decisions
- `pattern` — recurring code patterns in your project
- `error` — bugs you hit and how you fixed them
- `context` — project-specific rules and constraints
Memory injects automatically before each tool call. You explain things once. SkillGod remembers them.
Pro Vault
The free tier includes 30 curated starter skills. Pro unlocks the full vault of 1,944 skills.
After purchasing at skillgod.dev/#pricing, you receive a license key via email. Run:
```
sg sync --key YOUR_LICENSE_KEY
```
The vault syncs to `~/.skillgod/vault/`. Skills decrypt on your machine using your license key and machine ID combined. Nothing is sent to any server after sync.
MCP Server
SkillGod exposes a local MCP server on port 3333. It connects automatically with Claude Code, Cursor, and any MCP-compatible tool.
**Available MCP tools:**
- `find_skills(task)` — scored skill list
- `inject_context(task)` — augmented prompt string
- `save_memory(summary, kind)` — store a memory entry
- `get_memory(project, limit)` — retrieve project memory
- `stocktake()` — vault health report
- `vault_stats()` — skill counts and top skills