使用说明
SessionStellar JS
Open source packages for scoring AI orchestration quality. Parse session transcripts, compute scores across 5 metrics, and integrate into CI/CD pipelines.
Packages
| Package | npm | Description |
|---|---|---|
@sessionstellar/core |
Scoring engine & parser — zero network deps | |
sessionstellar |
CLI — terminal, CI/CD, git hooks | |
@sessionstellar/mcp |
MCP server — Claude Code, Cursor, Windsurf |
Quick Start
CLI
npx sessionstellar score session.md
MCP Server
# Claude Code
claude mcp add sessionstellar -- npx -y @sessionstellar/mcp
# Cursor — add to .cursor/mcp.json
{
"mcpServers": {
"sessionstellar": {
"command": "npx",
"args": ["-y", "@sessionstellar/mcp"]
}
}
}
Programmatic
import { parseSessionFile, ScoringEngine } from '@sessionstellar/core';
const signals = parseSessionFile(content, 'session.md');
const score = await ScoringEngine.score(signals, crypto.randomUUID());
console.log(score.overallScore); // 0–100
Git Hook (auto-score on commit)
npx sessionstellar enable
Metrics
| Metric | Weight | What it measures |
|---|---|---|
| Skill Diversity | 20% | Range of tools/skills used relative to task complexity |
| Decision Depth | 25% | Quality and quantity of architectural decisions |
| Error Recovery | 20% | How errors are caught and resolved |
| Compound Learning | 20% | Patterns recognized and applied across the session |
| Orchestration Mastery | 15% | Effective use of agents and sub-processes |
Development
pnpm install
pnpm run build
pnpm run typecheck
See CONTRIBUTING.md for details.
Related
- sessionstellar.com — Web app with leaderboard and analytics
- sessionstellar-cursor — Cursor IDE plugin
License
MIT