Drop this folder into any project, open Claude Code, and get a full forensic learning diagnostic in 60 seconds.
Actual report generated from a public repo — 207 commits, 3 days, 7 behavioral eras detected. Every chart cites real commit hashes.
You know that feeling when you've been coding for months and can't tell if you're actually getting better?
This tool reads your git history and tells you exactly what you learned, what you're missing, and what to study next. Every claim cites a commit hash from your actual repo. No setup. No data entry. No subjective guesses.
An auto-opening HTML report with three sections:
| Output | The Question | What You Actually Get |
|---|---|---|
| What You Learned | "Am I improving?" | Chronological narrative with velocity metrics, behavioral eras, breakthrough detection |
| What You're Missing | "What's holding me back?" | Ranked knowledge gaps backed by behavioral evidence — frustration patterns, rework hotspots, blind spots |
| What to Study Next | "What should I learn?" | ROI-ranked curriculum with hands-on exercises and real video recommendations from verified creators |
The report includes 8 interactive visualizations: era timelines, velocity curves, heatmaps, gap severity donuts, rework bars, and a curriculum roadmap — all in a single self-contained HTML file with zero dependencies.
git clone https://github.com/KyaniteLabs/dev-learning-archaeologist.git
cp -r dev-learning-archaeologist /path/to/your-project/
cd /path/to/your-project && claudeThen paste:
Analyze this repository's git history using the Dev Learning Archaeologist
methodology. Start with Phase 0 (ground truth), then proceed through all 5 phases.
That's it. The report opens in your browser automatically.
The archaeologist runs a 5-phase forensic pipeline on your repo:
- Ground Truth — Count commits, consolidate identities, establish baseline metrics
- Excavate — Extract commit types, temporal patterns, burst-gap cycles, file hotspots
- Stratify — Detect behavioral eras by velocity shifts, intent changes, and technology adoption
- Analyze — Run 7 independent analysis vectors in parallel
- Deliver — Generate a self-contained HTML report and open it in your browser
| # | Vector | What It Finds |
|---|---|---|
| 1 | Learning Velocity | How fast you're learning new concepts, and whether it's accelerating |
| 2 | Frustration Detection | Files you keep revisiting, fix clusters, where you're stuck (not just iterating) |
| 3 | AI Collaboration Maturity | Your autonomy level (L1 Directed → L4 Supervisory) and trust trajectory |
| 4 | Knowledge Gaps | Reinvented wheels, missing fundamentals, and what's causing rework |
| 5 | Temporal Behavior | Peak creative hour, optimal work patterns, burst sustainability |
| 6 | Cross-Domain Transfer | Skills from non-coding domains showing up in your code |
| 7 | External Learning | YouTube watch history → commit correlation (with Google Takeout) |
Every finding cites a commit hash. Every recommendation traces back to evidence.
| Data Source | Where It Looks | Required? |
|---|---|---|
| Git history | .git/ in the current project |
Yes — this is the minimum |
| Session logs | .claude/ directory (Claude Code), .cursor/ or Copilot exports |
Optional — unlocks AI maturity scoring |
| Cross-repo history | Other local repos you point to | Optional — unlocks cross-domain transfer |
| YouTube history | data/ folder (Google Takeout JSON) |
Optional — unlocks learning latency measurement |
It works with git history alone. Everything else makes the analysis richer, but git is the only requirement.
This is an Interpretable Context Methodology specialist — folder structure as agent architecture. Each file has one job:
| File | Job |
|---|---|
identity.md |
Who the specialist is — loads first |
rules.md |
The 5-phase pipeline, 7 vectors, output constraints |
examples.md |
Conversational demos showing the specialist in action |
reference/signal-heuristics.md |
Era classification, frustration levels, formulas |
reference/output-schemas.md |
JSON schemas for structured outputs |
reference/html-report-spec.md |
Design system — dark theme, 8 chart types, responsive |
reference/verified-creators.md |
Five trusted creators for learning plan recommendations |
reference/data-enrichment.md |
Google Takeout setup, supported data sources |
This is the lightweight diagnostic — zero install, runs in a Claude Code conversation. A full forensic pipeline is in the works: SQLite databases, Datasette inspection, multi-project sync, automated audits, and 20+ CLI commands. Coming soon from KyaniteLabs.
| Learning Archaeologist | Full Pipeline (coming soon) | |
|---|---|---|
| Setup | Drop in a folder | pip install |
| Runs in | Claude Code conversation | CLI / Python API |
| Vectors | 7 learning-focused | 6 + 14 opportunity analyzers |
| Output | HTML report | HTML + SQLite + Datasette + Markdown |
| Best for | "How am I doing?" | "Archaeologically analyze everything" |
Simon Gonzalez de Cruz — KyaniteLabs. We build AI-native developer tools.
MIT