Turn your AI coding sessions into skill growth
Sparkey Reflect analyzes how you use Claude Code and generates personalized coaching insights grounded in DORA, SPACE, DevEx, GitClear, and METR research benchmarks. All analysis runs locally — no data leaves your machine.
Industry-benchmarked analysis
Smooth, continuous scoring curves across every aspect of AI-assisted development — not step-function grades, but signals that reflect gradual improvement.
Specificity, context richness, clarity, efficiency, chain of thought
GitClear: specific prompts → 40% less churn
Turns to resolution, correction rate, context retention, iteration velocity
DORA: fewer iterations = faster lead time
File references, error context, code snippets, scope clarity
METR: code context → better completions
Duration, frequency, diversity, fatigue detection, deep work alignment
DORA 2024: uninterrupted blocks boost throughput
Tool diversity, MCP utilization, slash commands, automation
Specialized tools = mastery signal
Completeness, specificity, actionability, currency, ecosystem coverage
DORA: stale docs = risk
AI commit rate, productivity, rework rate, quality signals
GitClear 2024: AI rework benchmarks
Scoring grounded in industry-standard frameworks (DORA, SPACE, DevEx) and validated against published research from GitClear and METR.
DORA is a program of Google Cloud. GitClear and METR are independent organizations. Sparkey is not affiliated with or endorsed by any of these entities.
Analyze. Dive deep. Improve.
Report Mode
/sparkey:reflectRun daily, weekly, monthly, or full analysis to get severity-ranked insights with real session evidence and next-step recommendations.
Deep Dive Mode
/sparkey:reflect deep-dive <skill>Get an in-depth analysis of one skill area with before/after examples and tailored practice exercises.
Update Rules Mode
/sparkey:reflect update-rulesAutomatically improve your CLAUDE.md based on real session data — close the gap between how you work and what your AI knows.
Real-world impact
Weekly Skill Check-In
A senior developer runs /sparkey:reflect every Monday to track AI coding effectiveness. They discover a 23% correction rate — nearly 1 in 4 AI responses need fixing. The deep-dive reveals they skip error context when debugging, forcing extra back-and-forth. After two weeks of including stack traces, their correction rate drops to 8%, saving ~30 minutes per day.
Onboarding Teams to AI-Assisted Development
An engineering manager installs the plugin for a team of 6 developers. Junior developers run /sparkey:reflect monthly for a comprehensive baseline. Reports highlight specific patterns — one developer uses Bash(sed) instead of Edit, another has 3-hour marathon sessions with declining quality. After one quarter, the team's average score improves from 48 to 71.
Improving Project Rule Files
A tech lead runs /sparkey:reflect update-rules to optimize their CLAUDE.md. The analysis reveals only 1 instruction file with no code examples. After the update, First Response Acceptance jumps from 52% to 74% because Claude follows project conventions on the first try.
From sessions to insights
Up and running in 30 seconds
Install via Plugin Marketplace
Requires Claude Code 1.0.33+ and Python 3.11+
Install CLI Standalone
Optional — use outside of Claude Code
Quick Start
Free and open source
MIT License — forever
- ✓Full scoring engine (7 dimensions, 35 sub-dimensions)
- ✓CLI analysis (sparkey-reflect analyze)
- ✓Claude Code plugin (/sparkey:reflect)
- ✓Local SQLite trend storage
via sparkey.ai
- ✓Everything in Open Source
- ✓Version Control + Project Management integrations
- ✓Multi-developer trend tracking and comparison
- ✓Team-level usage analytics
- ✓Industry benchmark comparisons
- ✓Manager dashboard
Built for privacy and speed
Ready to level up your AI coding skills?
Install Sparkey Reflect — takes 30 seconds, runs entirely on your machine.
