What It Does — A TDD-First Skill Suite for Claude Code
Upstack is a lightweight skill suite for Claude Code designed specifically for red/green test-driven development workflows. Created by Upsolve AI's CTO Serguei (formerly of Palantir's Hyperauto), it addresses a gap in the current Claude Code ecosystem: while tools like gstack excel at greenfield projects, Upstack focuses on the "last-mile" iterations where testing, correctness, and polish matter most.
The suite includes 9 focused skills that work with any agent supporting the SKILL.md standard, including Claude Code, Cursor, and Gemini CLI. The skills are deliberately compatible with gstack, so you can use both simultaneously.
Setup — 30-Second Installation
Installation is straightforward:
git clone https://github.com/Upsolve-Labs/upstack.git ~/.claude/skills/upstack
cd ~/.claude/skills/upstack && ./install.sh
The install script handles linking skills and provides an INSTALL_STATUS report with NEXT_STEPS. For manual installation with other tools:
# For Claude Code
cp -r ~/.claude/skills/upstack/skills/* .agents/skills/
# For Cursor
cp -r ~/.claude/skills/upstack/skills/* .cursor/skills/
# For Gemini CLI
cp -r ~/.claude/skills/upstack/skills/* .gemini/skills/
Skills are plain markdown files, making them portable across any agent that reads SKILL.md format.
When To Use It — The TDD Workflow in Action
Upstack shines when you need to iterate on existing codebases with confidence. The workflow follows a strict TDD pattern:
- Start with
/advisor— Assess your current codebase status - Plan with
/plan— For a feature like "telemetry dashboard for my SaaS app," Claude will audit your codebase, compare alternatives, map error paths, create test coverage diagrams, and generate structured tickets with dependency DAGs - Execute with
/execute— Claude navigates your app with agent-browser, writes failing tests first (red), implements functionality, then confirms tests pass (green) - Ship with
/ship-pr— Open a PR complete with screenshots and evidence
You can also execute specific tickets: /execute --ticket P1-3 will read the ticket brief, check dependencies, scope tests to acceptance criteria, implement, and mark as done in TODOS.md.
The /validate skill walks through every planned path manually, screenshots UI, and saves evidence — perfect for security-focused teams like Upsolve that require auditable data and reliability.
Why This Matters Now
With Claude Code making code "essentially free to write," the bottleneck shifts to specification, testing, and correctness. Upstack forces the discipline of thinking test-first before touching any code. This aligns with recent Claude Code developments around subagent features for isolated task execution and the growing MCP ecosystem for domain-specific expertise.
Teams building data agents or working on security-sensitive applications will particularly benefit from Upstack's focus on provable, reliable code with real evidence. The screenshot and Postman collection generation gives you confidence to ship PRs to production faster.
Getting Started Today
Try Upstack on your next bug fix or small feature. The 30-second setup means you can test it immediately. Start with /advisor to see where your project stands, then use /plan on a well-scoped task. Notice how the TDD approach changes your interaction with Claude Code — you're not just asking for code, you're asking for provable, tested functionality with built-in validation.
Since skills are compatible, you can keep using gstack for larger architectural work while applying Upstack's TDD rigor to the implementation details.
