oh-my-prompt
Stop guessing what works. Let your prompts tell you.
English |
한국어
A Claude Code plugin that turns your own usage data into actionable improvements. It quietly logs every prompt, token, tool call, and session into a local SQLite database, then helps you:
- Surface short prompts you keep retyping → candidates for
CLAUDE.md rules, memory entries, or new slash commands
- Compare your high-leverage sessions against your low-leverage ones to learn what kind of opening prompt actually works for you
- Hand the raw analysis to a sub-agent and let it draft the patches
100% local. No outbound calls. No API key required (uses your existing claude CLI auth).
Quick start
/plugin marketplace add github.com/handlecusion/oh-my-prompt
/plugin install omp@oh-my-prompt
The UserPromptSubmit and Stop hooks register themselves and start writing to ~/.claude/omp.db. After a session or two:
/omp:dashboard
That single tabbed page gives you Stats, Patterns, Efficiency, and your Suggest archive — all in one window.
Slash commands
| Command | Args | What it does |
|---|
/omp:stats [days] | days (default 7) | Daily token usage, per-model breakdown, per-project distribution, session stats — rendered as a web dashboard |
/omp:patterns [days] [min] | 30, 3 | Repeated prompts, intent distribution, first-4-words patterns, CLAUDE.md candidates |
/omp:efficiency [days] [min] | 30, 3 | Per-session leverage / autonomy / tool metrics, top vs bottom 20% comparison, first-prompt analysis |
/omp:suggest [days] [min] | 30, 3 | Runs both analyses, hands the raw output to an Opus sub-agent, and writes a structured proposal (rules, memory, slash candidates) to ~/.claude/omp_suggestions/<timestamp>.md |
/omp:suggest-archive | — | Browse the cumulative ~/.claude/omp_suggestions/ archive in a sidebar viewer |
/omp:dashboard [stats_days] [days] [min] | 7, 30, 3 | All four panels above in a single tabbed page (1-4 / j/k / arrows to switch) |
Screenshots
/omp:dashboard — one tabbed page that wraps the four panels below:
/omp:patterns — repeated prompts, intent distribution, first-4-words patterns, auto suggestions:
/omp:efficiency — per-session leverage / autonomy, top-vs-bottom 20% comparison, side-by-side first-prompt cards, tool usage delta:
/omp:suggest-archive — sidebar list + DOMPurify-rendered markdown of every cumulative /omp:suggest analysis:
Screenshots above are generated from scripts/build_mock_db.py --locale en — fully synthetic data so no real user prompt is exposed. The dashboard chrome is also swapped to English via scripts/localize_html.py before capture (the runtime UI is Korean).
Install (option 2: clone manually)
git clone https://github.com/handlecusion/oh-my-prompt ~/Code/oh-my-prompt
Add to the hooks block of ~/.claude/settings.json:
