tinyspec vs GSD
GSD (Get Shit Done) is a meta-prompting and context engineering framework for Claude Code, OpenCode, Gemini CLI, and Codex. It orchestrates spec-driven development with parallel multi-agent execution, designed to combat context window degradation across long sessions.
Overview
| tinyspec | GSD | |
|---|---|---|
| Approach | Spec authoring tool | Multi-agent orchestration |
| Spec format | Single Markdown file | XML tasks + generated Markdown in .planning/ |
| CLI | Native binary, full spec management | No standalone CLI (slash commands inside LLM tools) |
| Install | Single binary | npm (npx get-shit-done-cc@latest) |
| LLM dependency | Optional | Required (slash commands inside LLM tools) |
Spec format
GSD generates planning documents in a .planning/ directory plus top-level files (PROJECT.md, REQUIREMENTS.md, ROADMAP.md, STATE.md). Individual tasks use an XML structure optimized for LLM consumption. Files are organized by phase number and deliberately kept small to stay under LLM context quality thresholds.
tinyspec uses a single Markdown file per feature. The format is human-readable and human-writable, with a clear four-section structure. Plain Markdown, no XML, no generated file trees.
CLI tooling
GSD has no standalone CLI. It installs as slash commands within LLM tools (Claude Code, OpenCode, Gemini CLI, Codex). Commands like new-project, plan-phase, execute-phase, and verify-work run inside an active AI session.
tinyspec provides a standalone binary with commands for the full spec lifecycle. Every command works from the terminal without any AI agent running.
Workflow integration
GSD orchestrates multi-agent execution with parallel sub-agents for research, planning, implementation, and verification. Tasks are grouped into dependency waves — independent tasks run in parallel, waves run sequentially. Each task gets a fresh 200k-token context. Model profiles (Quality, Balanced, Budget) control cost.
tinyspec integrates with Claude Code via skills. You control the workflow: refine a spec, then implement it task by task. Specs are the artifact, not an intermediate step in an agent pipeline.
Formatting
GSD has no formatting tooling. Planning documents are generated programmatically through LLM prompts.
tinyspec format normalizes your Markdown so specs stay consistent across authors and edits.
When to choose GSD
- You want automated multi-agent execution with parallel task processing
- Context window management is a priority for your large projects
- You need wave-based dependency resolution for complex implementation plans
- You want to control cost with model profile selection
When to choose tinyspec
- You want a spec authoring tool, not an execution orchestrator
- You prefer human-readable specs over LLM-optimized XML tasks
- You need a standalone CLI that works without an AI agent
- You want minimal project overhead — one
.specs/directory vs. a generated.planning/tree