Get Shit Indexed

A light-weight and powerful meta-prompting, context engineering and spec-driven development system for Claude Code, OpenCode, and Gemini CLI.

Solves context rot — the quality degradation that happens as Claude fills its context window.

npx get-shit-indexed-cc@latest

Why This Fork

Forked from get-shit-done with MCP tool integration for 80-90% token savings.

Native tools (Bash, Read, Write, Grep, Glob) burn through context windows with protocol overhead. GSI uses MCP tools instead:

| Operation | Native | MCP | Savings | |-----------|--------|-----|---------| | Read file | ~15K tokens | ~2K tokens | 87% | | Write file | ~15K tokens | ~2K tokens | 87% | | Search code | ~15K tokens | ~3K tokens | 80% | | List files | ~15K tokens | ~2K tokens | 87% |

Same work, 80-90% less context waste.

Quick Start

# Interactive install
npx get-shit-indexed-cc

# Non-interactive
npx get-shit-indexed-cc --claude --global
npx get-shit-indexed-cc --opencode --global
npx get-shit-indexed-cc --gemini --global
npx get-shit-indexed-cc --all --global

Verify with /gsi:help inside Claude Code.

Features

Prompt Enhancement System (NEW)

GSI now includes an intelligent prompt analysis and enhancement system:

• Risk Assessment — Scores prompts 0-100 for complexity
• Mode Selection — Auto-selects enhancement intensity (none/light/standard/full)
• Template System — 5 enhancement templates (Engineering, Clarity, Decomposed, Academic, Security)
• Local Processing — No external API calls, <5ms response time

const { analyzePrompt, fullEnhance } = require('get-shit-indexed-cc/lib/prompt-enhancer');

const analysis = analyzePrompt('Implement authentication');
// { riskScore: 45, mode: 'standard', template: 'engineering' }

const result = fullEnhance('Build a REST API');
// Enhanced prompt with structure and clarity improvements

Thinking Servers

Three thinking servers for enhanced reasoning:

• Sequential — Multi-step problem decomposition
• Tractatus — Logical structure analysis
• Debug — Graph-based problem-solving

GSI Statusline v2.0

Tokyo Night themed statusline with:

• Phase detection from STATE.md
• 8-segment gradient progress bar
• Context usage display
• Model identification

How It Works

1. /gsi:new-project — Initialize project with research and requirements
2. /gsi:plan-phase N — Research + plan + verify phase N
3. /gsi:execute-phase N — Execute plans in parallel waves
4. /gsi:verify-work — User acceptance testing
5. /gsi:complete-milestone — Archive and release

MCP Servers

GSI uses two MCP servers for token-efficient operations:

Desktop Commander — File operations, process management

• read_file, write_file, list_directory, start_process
• 80-90% token savings vs native tools

Code-Index MCP — Code search and symbol navigation

• search_code_advanced, find_files, get_symbol_body
• 70-80% token savings vs Grep

Commands

| Command | Purpose | |---------|---------| | /gsi:new-project | Initialize project | | /gsi:plan-phase N | Plan phase N | | /gsi:execute-phase N | Execute phase N | | /gsi:verify-work | UAT validation | | /gsi:progress | Show status | | /gsi:help | All commands | | /gsi:yolo | Maximum speed mode |

Project Structure

get-shit-indexed/
├── bin/              # CLI installer
├── commands/         # 26 gsi commands
├── agents/           # 12 gsi agents
├── workflows/        # 30+ workflow definitions
├── references/       # Documentation
├── templates/        # Plan templates
├── lib/              # Core modules
│   ├── prompt-enhancer/  # Risk assessment + enhancement
│   ├── complexity/       # Complexity prediction
│   ├── thinking/         # Thinking orchestration
│   └── workflow-thinking/# Workflow integration
└── .planning/        # Project state

Recent Changes

v1.22.0 (2026-02-16)

• Prompt Enhancement System — Risk assessment engine with 0-100 scoring
• 5 enhancement templates (Engineering, Clarity, Decomposed, Academic, Security)
• Mode selection (none/light/standard/full) based on prompt complexity
• Local-only processing, <5ms response time

v1.21.0 (2026-02-16)

• GSI Statusline v2.0 — Tokyo Night edition
• CodeGraphContext removed — 2-server architecture (DC + CI only)
• Git author rewrite — all commits show "Mose" as author

Credits

Forked from get-shit-done by TÂCHES.

Original GSD provided workflow structure and planning methodology. GSI adds:

• MCP tool integration (80-90% token savings)
• Three thinking servers
• Prompt enhancement system
• Simplified 2-server architecture

License

MIT