Live Developer Tools AI Open Source

sp3cmar

Structured development framework for AI coding agents

A skill library and review framework that brings discipline to AI-assisted development — specs before code, reviews before merge, accountability throughout.

Visit product Source code

The problem

AI coding agents are powerful but undisciplined. Without structure, you get “vibe coding” — the AI writes something, you deploy it, and problems surface later. The gap between what AI agents can do and what disciplined software engineering requires is widening.

What sp3cmar does

sp3cmar brings the discipline. It’s a framework of Claude Code skills that encode a structured development workflow: write a spec before touching code, review at multiple stages before merging, and track accountability across sessions.

The result is AI-assisted development that’s auditable, repeatable, and safe enough for production.

Components

Skills library (20+ skills)

Each skill encodes a repeatable workflow as a Claude Code command:

  • /sp3cmar-feature — write a spec with outcomes and acceptance criteria
  • /sp3cmar-implement — autonomous implementation from spec to working PR
  • /sp3cmar-review-pr — standards review with ship-stopper identification
  • /sp3cmar-review-kill — adversarial “kill case” review
  • /sp3cmar-review-codebase — architecture review with delta tracking
  • /sp3cmar-ship — lint, commit, push, and create PR in one step
  • /sp3cmar-morning — morning session startup with context and priorities

Review framework

Three review stages catch different classes of problems:

  1. Architecture review (before building) — is this the right approach?
  2. Correctness review (before merging) — does it work and meet the spec?
  3. Kill-case review (before shipping) — what could go wrong in production?

Session management

Morning briefings pull in recent commits, open PRs, and pending work. Session debriefs capture what was done and persist context for the next session — even when switching between AI providers.

Under the hood

  • Skills: Markdown prompt files with YAML frontmatter, installed as Claude Code commands
  • Architecture: Copied from sp3cmar into GTM:OS with domain-specific adaptations
  • CI: GitHub Actions for linting, type checking, and test runs

Status

Publicly released and open source. In daily use across all B3dmar products. The skill framework was recently generalized into GTM:OS for go-to-market workflows.

All products