reprompt-cli 0.3.2

Discover, analyze, and evolve your best prompts from AI coding sessions

reprompt

Discover, analyze, and evolve your best prompts from AI coding sessions.

repomix packs your code for AI. reprompt extracts insights from AI.

Every developer's AI session history contains reusable prompt patterns -- scattered across hundreds of session files. reprompt extracts them, deduplicates, analyzes frequency, and builds a personal prompt library that evolves over time.

Quick Start

pipx install reprompt-cli
reprompt scan
reprompt report
reprompt library

Terminal Report

reprompt -- AI Session Analytics
========================================

 Overview
 Total prompts:        1,247
 Unique (deduped):       832
 Sessions scanned:       156
 Sources: claude-code, openclaw

 Top Prompt Patterns
 #  | Pattern                  | Count | Category
 1  | fix the failing test...  |    42 | debug
 2  | add unit tests for...    |    38 | test
 3  | refactor X to use...     |    27 | refactor

Features

• Auto-detection -- finds Claude Code and OpenClaw sessions automatically
• Two-layer dedup -- SHA-256 exact + TF-IDF semantic similarity
• Hot terms analysis -- TF-IDF discovers your most-used technical terms
• K-means clustering -- groups similar prompts into themes
• Prompt library -- extracts high-frequency patterns, auto-categorizes (debug/implement/test/review/refactor/explain/config)
• Rich reports -- beautiful terminal output with tables and bar charts
• Multiple formats -- terminal, JSON (for pipelines), Markdown (for docs)
• Pluggable adapters -- add support for any AI coding tool
• Prompt search -- find past prompts by keyword across all sessions
• Zero config -- works out of the box, customize via env vars or TOML

How reprompt Compares

Feature	reprompt	prompt-manager	agent-sessions	cclog
Multi-tool support	✅ Claude, OpenClaw, + adapters	✅ Multiple	✅ Multiple	❌ Claude only
Exact dedup (SHA-256)	✅	❌	❌	❌
Semantic dedup (TF-IDF)	✅	❌	❌	❌
Hot terms analysis	✅ TF-IDF	❌	❌	❌
K-means clustering	✅	❌	❌	❌
Pattern library	✅ Auto-categorized	❌	❌	❌
CLI interface	✅	✅ TUI	❌ macOS app	✅
JSON/Markdown export	✅	❌	❌	❌
Pluggable adapters	✅	✅	❌	❌
Zero config	✅	✅	✅	✅

Supported AI Tools

Tool	Status	Session Path
Claude Code	Supported	~/.claude/projects/
OpenClaw / OpenCode	Supported	~/.openclaw/ + ~/.opencode/sessions/
Codex CLI	Planned (v0.4)	~/.codex/
Aider	Planned	~/.aider/
Gemini CLI	Planned	~/.gemini/
Continue.dev	Via MCP	MCP protocol
Zed	Via MCP	MCP protocol
Cursor	Planned	--

Usage

# Scan all detected AI tools
reprompt scan

# Scan specific source
reprompt scan --source claude-code

# Scan custom path
reprompt scan --path ~/custom/sessions

# Rich terminal report
reprompt report

# JSON output (for CI/pipelines)
reprompt report --format json

# Search your prompt history
reprompt search "authentication"
reprompt search "debug" --limit 5

# View your prompt library
reprompt library

# Filter by category
reprompt library --category debug

# Export prompt library as Markdown
reprompt library prompts.md

# Database stats
reprompt status

# Auto-scan after sessions
reprompt install-hook

# Cleanup old data
reprompt purge --older-than 90d

MCP Server (Claude Code / Continue.dev / Zed)

reprompt includes an MCP server that exposes your prompt analytics as tools for AI coding assistants.

pip install reprompt-cli[mcp]

# Start the server
reprompt mcp-serve

Register in Claude Code — add to .mcp.json at project root:

{
  "mcpServers": {
    "reprompt": {
      "type": "stdio",
      "command": "reprompt",
      "args": ["mcp-serve"]
    }
  }
}

Available tools: search_prompts, get_prompt_library, get_best_prompts, get_trends, get_status, scan_sessions

Available resources: reprompt://status, reprompt://library

Configuration

Zero config by default. Customize with environment variables or TOML:

# Environment variables (prefix: REPROMPT_)
REPROMPT_EMBEDDING_BACKEND=ollama reprompt scan
REPROMPT_DB_PATH=~/custom/reprompt.db reprompt status

# ~/.config/reprompt/config.toml
[embedding]
backend = "tfidf"  # tfidf | ollama | local | openai

[storage]
db_path = "~/.local/share/reprompt/reprompt.db"

[dedup]
semantic_threshold = 0.85

[library]
min_frequency = 3

Optional Backends

pip install reprompt-cli[ollama]   # Ollama API embeddings
pip install reprompt-cli[local]    # sentence-transformers (CPU)
pip install reprompt-cli[openai]   # OpenAI API embeddings

Adding an Adapter

Create a new adapter by subclassing BaseAdapter:

from reprompt.adapters.base import BaseAdapter
from reprompt.core.models import Prompt

class MyToolAdapter(BaseAdapter):
    name = "my-tool"
    default_session_path = "~/.my-tool/sessions"

    def parse_session(self, path):
        # Parse session file -> list[Prompt]
        ...

    def detect_installed(self):
        return Path(self.default_session_path).expanduser().exists()

Troubleshooting

NumPy conflict in Anaconda environments

If you see an error like:

A module that was compiled using NumPy 1.x cannot be run in NumPy 2.x

This happens when Anaconda's base environment has packages compiled against NumPy 1.x but a newer NumPy 2.x is installed. This is not a reprompt bug — it's an environment conflict.

Fix: Install reprompt in an isolated environment using pipx:

pip3 install --user pipx
pipx install reprompt-cli
reprompt scan

pipx creates a dedicated virtualenv for reprompt, avoiding conflicts with your system Python or Anaconda.

Roadmap

• Prompt version control — track how your prompts evolve across iterations, with semantic diffing and per-version effectiveness scoring
• More adapters — Codex CLI, Aider, Gemini CLI, Cursor
• Team analytics — aggregate insights across team members (opt-in, anonymized)
• Prompt recommendations — suggest better prompts based on your history and outcomes

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

MIT