Stay organized and in control with adaptive LLM workflow prompts

A single rendered entry point dynamically assembles LLM instructions from 17 development modes (planning, setup, coding, debugging, documentation, refactoring, and more) using Jinja2 templates. Type "go" in the LLM chat to advance through each step. Switch modes in the CLI as needed to get the right guidance for the task at hand.

Works for any programming language or framework. Integrated with Pyve to manage project environments.

This is "HITLoop" (human-in-the-loop) development: you direct, the LLM executes. You stay in charge, guiding features, flow, and taste, while the LLM handles the mundane coding. The pace is "flaming agile"—an entire production-ready backend can be completed in 6-12 hours. When you customize a prompt for your project, mark it as overridden so future package updates skip it. When you want the latest workflow improvements, a simple command will sync all non-overridden prompts.

View on GitHub Install from PyPI Get Started

Quick Start

1. Install

Install via pip or pipx for system-wide CLI access

pip install project-guide

2. Initialize

Navigate to your project directory and initialize

project-guide init

3. Start

Tell your LLM to read the rendered entry point

Read `docs/project-guide/go.md` and start.

4. Advance

Type "go" to advance through each workflow step

go

5. Switch Modes

Change modes to match your current task

project-guide mode <mode-name>

6. Update

Pull latest workflow improvements

project-guide update

7. Check Status

See which files are current, outdated, or locked

project-guide status

Features

Core Capabilities

Mode-Driven Workflows

17 development modes—planning, setup, coding, debugging, documentation, refactoring, and more—each tailored with the right instructions for the task.

Pyve Integration

First-class support for pyve-managed hosting: install globally with pyve self install, generate a named-environment spec with the plan_envs mode, and get pyve-aware onboarding plus a status host footer and heal drift warnings when pyve manages the install.

Dynamic Rendering

Jinja2 templates assemble a single rendered entry point (go.md) from mode-specific content, so the LLM always sees exactly what it needs.

Hash-Based Sync

Content-hash tracking detects changes precisely, so updates only touch files that actually differ from the package version.

File Override Lock

Lock any customized file to prevent future package updates from overwriting your project-specific changes.

Compact Status

Color-coded status output shows at a glance which files are current, outdated, locked, or missing.

Zero Configuration

Works out of the box with sensible defaults. Run init and start working immediately.

Operational Benefits

Cross-Platform

Runs on macOS, Linux, and Windows with Python 3.11+ for consistent workflows everywhere.

Well Tested

596 tests with 90% coverage ensure reliability across platforms and edge cases.

Lightweight and Fast

Minimal dependencies (click, jinja2, pyyaml, packaging) mean fast installation and no bloat.

Selective File Updates

Use the --files flag to update specific files, or update everything at once with a single command.

Shell Completion

Tab completion for commands, flags, and mode names in bash, zsh, and fish — including dynamic completion of mode names from your project's metadata.

Safe Operations

Idempotent commands and explicit consent requirements protect your project-specific customizations.

Gentle Forced Updates

Automatic .bak files created if you force-update a locked file, so nothing is lost.

Development Philosophy

HITLoop Development

You direct features, flow, and taste. The LLM handles the typing. Human-in-the-loop collaboration at its best.

Flaming Agile Pace

Complete an entire production-ready backend in 6-12 hours with structured, methodical LLM collaboration.

Structured Workflow

The workflow enforces a focused methodology across 17 modes with approval gates at every step.