A Claude skill that tracks your coding sessions and tells you exactly how you could have prompted better — saving you time, back-and-forth, and frustration.
Most developers lose 30–60% of a session to corrections that could've been avoided with a clearer initial prompt. Session Analyzer fixes that.
How it works:
- Type
init-session→ Claude starts silently tracking - Work normally — describe your task, share files, get code, make corrections
- Type
finish-session→ Claude analyzes the entire session and generates 4 report files
| File | What's in it |
|---|---|
session-rating.md |
Score out of 10 across 4 dimensions with a full breakdown |
better-prompt.md |
Your original prompt rewritten the way it should've been, with a diff table |
prompt-guide.md |
How to approach this type of task better in future sessions |
prompt-patterns.md |
Reusable prompt templates extracted from your session |
- Prompt Clarity & Completeness (30%) — Was the task explained well enough upfront?
- Correction Rounds (30%) — How many "no, fix this" messages were needed?
- Asset & File Organization (20%) — Were files/screenshots shared at the right time?
- Final Result Alignment (20%) — How close was the result to what you actually wanted?
- Download this repo as a ZIP (
Code → Download ZIP) - Go to Customize → Skills in Claude.ai
- Click + → Create skill → upload the ZIP
- Make sure Code Execution is enabled in Settings → Capabilities
Skills are available on Pro, Max, Team, and Enterprise plans.
git clone https://github.com/Zein0/claude-session-analyzer.git ~/.claude/skills/session-analyzerStart a new Claude Code session — the skill is active immediately.
You: init-session
Claude: ✅ Session started. Describe your task whenever you're ready.
You: build me a login form with email + password validation using React and Tailwind
[... you work, Claude codes, you correct things ...]
You: finish-session
Claude: Session ended. Analyzing now — generating your 4 report files...
Overall Score: 6.4/10
Prompt Clarity: 5/10 — No mention of validation library, error message style, or existing design system
Correction Rounds: 6/10 — 3 corrections, 2 of which were about styling that could've been specified upfront
Asset Organization: 7/10 — Design screenshot shared after first attempt instead of before
Final Alignment: 8/10 — End result matched what was needed
Original: "build me a login form with email + password validation"
Better: "Build a login form in React using Tailwind CSS. Fields: email + password.
Validation: email format check, password min 8 chars. Show inline error messages
in red below each field. On submit, call /api/auth/login. Match the style of the
existing Button and Input components in /components/ui/. Don't use any external
form libraries — plain React state is fine."
Estimated improvement: ~65% fewer correction rounds
The best way to get better at prompting AI is to see exactly where your prompts fell short — right after a session while it's fresh. Most people never do this, so they keep making the same mistakes.
This skill turns every session into a learning loop.
PRs welcome. If you have ideas for new scoring dimensions, report formats, or session patterns — open an issue.
MIT