CLI utility for analyzing CLAUDE.md files and generating structured prompts for Claude Code
Project description
Guru
CLI utility for analyzing CLAUDE.md files and generating structured prompts for Claude Code.
Guru reads your project's CLAUDE.md file and uses it to generate context-aware prompts that include role detection, relevant context filtering, tech stack, coding conventions, and project constraints.
Installation
# From PyPI (when published)
pip install talk-as-guru
# From source
git clone https://github.com/grigory-zhovtun/talk-as-guru.git
cd talk-as-guru
pip install -e .
Quick Start
- Create a
CLAUDE.mdfile in your project root:
# Tech Stack
- Python 3.11
- FastAPI
- PostgreSQL
# Conventions
- Use type hints everywhere
- Follow PEP 8 style guide
- Write docstrings for public functions
# Rules
- Never commit secrets
- Always write tests for new features
- Run Guru:
guru "create a REST API endpoint for user authentication"
- Result — a structured prompt in
./prompts/prompt_01.md:
# Роль
Senior Backend разработчик (FastAPI, Python, PostgreSQL)
# Задача
create a REST API endpoint for user authentication
# Техническое окружение
- Python 3.11
- FastAPI
- PostgreSQL
# Релевантный контекст
[Only relevant sections from CLAUDE.md]
# Конвенции
- Use type hints everywhere
- Follow PEP 8 style guide
# Ограничения и правила
- Never commit secrets
- Always write tests for new features
Two Modes
Standard Mode (default)
Generates a prompt locally using a template with:
- Automatic role detection — determines the role from query keywords and tech stack (e.g. "Senior Frontend разработчик (Next.js, React, TypeScript)")
- Relevant context filtering — includes only project-specific sections from CLAUDE.md, sorted by relevance to the query
- Structured output — role, task, tech environment, context, conventions, constraints
guru "add a notification bell to the header"
Smart Mode (--smart)
Uses Claude Code CLI to analyze the query and generate an enriched prompt with:
- AI-powered role selection — specific specialization for the task
- Task decomposition — 3-7 concrete subtasks with details
- Intelligent context filtering — only the most relevant parts of CLAUDE.md
- Technical constraints — derived from project architecture
guru --smart "add a notification bell to the header"
Requires Claude Code CLI installed (claude command in PATH).
Usage
Interactive REPL Mode
Start the interactive shell:
guru
Single Query Mode
Process a query without entering the REPL:
guru "create a function for email validation"
Command Line Options
guru --help Show help
guru --version Show version
guru --smart / -s Enrich prompt via Claude Code AI
guru -c /path/to/CLAUDE.md Use specific CLAUDE.md file
guru -p ./my-prompts Save prompts to custom directory
guru -t custom Use a custom template
guru --profile my-project Load a saved profile
REPL Commands
| Command | Description |
|---|---|
/reload |
Re-read CLAUDE.md from disk |
/history |
Show recent query history |
/template [name] |
Switch template (no arg = list available) |
/profile [name] |
Load profile (no arg = list available) |
/run |
Send last prompt to Claude Code CLI |
/smart [query] |
Enrich prompt via Claude Code AI |
/help |
Show help |
/exit |
Exit Guru |
Generated Prompts
Prompts are saved to ./prompts/ with auto-incrementing filenames:
prompts/
├── prompt_01.md
├── prompt_02.md
└── prompt_03.md
Each prompt includes:
- Automatically detected role
- Your query/task
- Tech stack from CLAUDE.md
- Relevant project context (filtered, not full dump)
- Project conventions
- Constraints and rules
Templates
Guru uses Jinja2 templates for prompt generation.
Available Template Variables
| Variable | Description |
|---|---|
user_query |
The user's query/task |
role |
Auto-detected role |
tech_stack |
Tech stack section |
relevant_context |
Filtered relevant sections |
conventions |
Conventions section |
constraints |
Constraints/rules section |
context |
Full file content (legacy) |
Custom Templates
Create custom templates in ~/.guru/templates/:
mkdir -p ~/.guru/templates
Example template (~/.guru/templates/detailed.md.j2):
# Role
{{ role }}
# Task Description
{{ user_query }}
## Technical Context
### Tech Stack
{{ tech_stack }}
### Relevant Context
{{ relevant_context }}
### Coding Standards
{{ conventions }}
### Important Rules
{{ constraints }}
Use it with:
guru -t detailed
# or in REPL
guru> /template detailed
Profiles
Save project configurations as profiles in ~/.guru/profiles/:
# ~/.guru/profiles/my-project.yaml
name: my-project
claude_md_path: ./docs/CLAUDE.md
template: detailed
prompts_dir: ./generated-prompts
Load with:
guru --profile my-project
# or in REPL
guru> /profile my-project
Claude Code Integration
/run — Execute prompt
Send the last generated prompt to Claude Code CLI:
guru> create a user model
Saved: ./prompts/prompt_01.md
guru> /run
Sending to Claude Code...
[Claude's response appears here]
/smart — AI-enriched prompts
Re-generate the last query with AI analysis:
guru> add notifications to the header
Saved: ./prompts/prompt_01.md
guru> /smart
Smart mode: enriching prompt via Claude Code...
Saved (smart): ./prompts/prompt_02.md
Or provide a new query directly:
guru> /smart add API endpoint for notifications
Requirements:
- Claude Code CLI must be installed and in your PATH
- Install from: https://claude.ai/code
CLAUDE.md Format
Guru recognizes sections based on keywords in headings:
| Section | Keywords |
|---|---|
| Tech Stack | tech, stack, technology, tools, dependencies |
| Conventions | convention, style, formatting, patterns, practices |
| Constraints | constraint, rule, restriction, important, warning |
All other sections are treated as project context and included based on relevance to the query.
Supported heading formats:
# Heading
## Heading
**Heading**
Role Detection
Guru automatically detects the appropriate role based on query keywords and tech stack:
| Keywords | Role |
|---|---|
| react, next.js, frontend, css, ui, компонент | Senior Frontend разработчик |
| python, fastapi, backend, api, endpoint | Senior Backend разработчик |
| sql, postgres, database, модел | Senior Database разработчик |
| docker, kubernetes, deploy, devops | Senior DevOps инженер |
| test, pytest, jest, e2e | Senior QA инженер |
| security, auth, jwt, oauth | Senior Security инженер |
| go, golang, goroutine | Senior Go разработчик |
The role is enhanced with specific technologies from the project's tech stack (e.g. "Senior Frontend разработчик (Next.js, React, TypeScript)").
Examples
Basic Workflow
# Navigate to your project
cd my-project
# Generate a prompt
guru "implement user authentication with JWT"
# Saved: ./prompts/prompt_01.md
# Generate an AI-enriched prompt
guru --smart "add rate limiting to the API"
# Saved (smart): ./prompts/prompt_02.md
REPL Workflow
guru
guru> implement user authentication with JWT
Saved: ./prompts/prompt_01.md
guru> /smart
Smart mode: enriching prompt via Claude Code...
Saved (smart): ./prompts/prompt_02.md
guru> /run
Sending to Claude Code...
guru> /history
guru> /exit
Configuration Directory
Guru stores configuration in ~/.guru/:
~/.guru/
├── templates/ # Custom Jinja2 templates
│ └── custom.md.j2
└── profiles/ # Project profiles
└── my-project.yaml
Development
# Install with dev dependencies
pip install -e ".[dev]"
# Run tests
pytest
# Format code
black src/
# Lint
ruff src/
License
MIT License - see LICENSE for details.
Contributing
Contributions are welcome! Please feel free to submit a Pull Request.
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file talk_as_guru-0.2.0.tar.gz.
File metadata
- Download URL: talk_as_guru-0.2.0.tar.gz
- Upload date:
- Size: 25.0 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.11.5
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
c7a1621d7b1bdaf7982a6935dcc2bc2319ddc52e97aafba6dcd271908e9b6f7e
|
|
| MD5 |
3350e33efa865afbed17c96a8c6583f3
|
|
| BLAKE2b-256 |
316773b1ea998d635fe5d87f590375a4ec766ca4d5eea172f7df6dec7e3605c5
|
File details
Details for the file talk_as_guru-0.2.0-py3-none-any.whl.
File metadata
- Download URL: talk_as_guru-0.2.0-py3-none-any.whl
- Upload date:
- Size: 23.8 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.11.5
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
96bded5138c5da264ed19c01e282766972a0f43cb15a14c216a4375bcf05d8b7
|
|
| MD5 |
e48687c5008a6492f3d828e0283492dc
|
|
| BLAKE2b-256 |
e65f00c629363d036a1b05319a0afad49e107dfb34f788d3b1caf830704ecad7
|