← all projects

Dev Tools Active 2025

get-shit-done

Meta-prompting system for AI coding

The problem

AI coding assistants are only as good as the context and structure they're given - without a living spec and consistent rules, every session re-derives the same project context and drifts from what was previously decided.

The approach

Built get-shit-done, an npm package that sets up intelligent context rules for a codebase in one command, maintains living specs for in-progress features, and orchestrates multi-step development workflows so an AI coding session picks up with full project context instead of starting cold.

TypeScriptNode.jsnpmContext Engineering

get-shit-done is a meta-prompting system for AI-assisted coding: instead of hand-crafting context every session, it builds and maintains that context for you, and orchestrates the multi-step workflows that turn a spec into shipped code.

one command, rich context

Running the setup command analyzes the codebase and installs intelligent context rules - the kind of project-specific knowledge (conventions, architecture, prior decisions) that would otherwise have to be re-explained to an AI assistant every session.

npx get-shit-done init
# scans the repo, generates context rules,
# sets up living spec tracking for in-progress features

living specs

Feature specs aren’t static documents - the workflow updates them as work progresses, so a spec reflects what’s actually been built, not just what was originally planned. That keeps context accurate across long-running, multi-session features.

workflow orchestration

Beyond context, the system orchestrates the multi-step development loop itself: breaking a feature into steps, tracking what’s done, and keeping the AI coding session anchored to the current spec rather than reinventing the plan each time.

distribution

Published on npm and cross-platform, so adopting it in a new project is a single install, not a bespoke setup process per repo.

status

Active: published and in use, with the core context-generation, spec-tracking, and workflow-orchestration pieces functional.

What I took away

  • Context engineering - what the model sees before it starts working - matters more than prompt wording for multi-step coding tasks.
  • Specs need to be living documents the workflow updates, not static files that go stale after the first session.
  • A single setup command is the difference between a tool people actually adopt and one they mean to try eventually.

0 comments

0/1000

Loading…