Skills & agents for the whole software lifecycle
A vendor-neutral, open-source developer plugin for Claude Code, GitHub Copilot, and Codex — requirements, architecture, design, implementation, review, testing, accessibility, debugging, and bounded autonomous repair.
34
Skills
20
Agents
10
Lifecycle groups
3
Runtimes
Features
Closed-loop by design: An EASE-MAS backbone links requirement → PRD → architecture → code → test → failure → root-cause → repair, with a supervisor orchestrating the DAG.
Independent critics: 15 specialist review agents (security, performance, a11y, tests, …) audit diffs read-only; a
deliberate step turns findings into a verdict.Bounded, safe autonomy: Repair is capped by
max_repair_attempts, guarded by a regression gate and an independent repair-validator. Merge stays human-owned.CLI-first tooling: Skills prefer real tools —
gh, az, npx playwright, npx axe — falling back to MCP servers when a CLI is unavailable.Accessibility built in: A full WCAG 2.1/2.2 AA lifecycle: scan, deep review, fix, verify, and generate Playwright + axe regression tests.
One config, three runtimes: A single
config.json plus prose companions drives Claude Code, Copilot, and Codex through thin generated adapters.Explore
Getting started
Install, configure, and invoke skills from any runtime.
Architecture
The canonical library, closed loop, traceability, and review pipeline.
Skills
All 34 lifecycle skills, generated from the real frontmatter.
Agents
20 agents — review critics and orchestrating leads.