mad-workflow

Determine the direct and transitive consequences of a MAD diff or controlled-contract change, classify compatibility and risk, mark stale work, and define proportionate revalidation. Use before QA/integration whenever shared artifacts or task premises may have changed.

Produce and validate canonical MAD completion, blocker, QA, impact, release, and responsibility-transfer reports with immutable revision evidence and explicit next ownership. Use whenever an agent finishes, pauses, fails, requests a gate, or transfers work.

Design and execute independent, risk-based validation for a pinned MAD task or integration candidate and issue a pass, pass-with-notes, fail, or blocked decision. Use for QA review, integration gates, regression analysis, and release readiness.

Implement an approved MAD task with minimal scope, established architecture, risk-driven validation, diff self-review, and truthful evidence. Use for backend, data, infrastructure, tooling, or mixed implementation work after task preflight.

Define, review, evolve, and register shared MAD contracts such as APIs, schemas, events, commands, configuration, persistence shapes, UI interfaces, migrations, and infrastructure boundaries. Use whenever work consumes or changes behavior shared across tasks or systems.

Implement or review frontend/mobile work using the product's approved design system, complete interaction states, semantic accessibility, responsive behavior, and evidence at the right layers. Use for any task that changes user-visible UI or interaction.

Create and verify isolated Git branches/worktrees for MAD tasks and integration, preserving ownership, base provenance, clean state, safe conflict handling, and cleanup. Use whenever agents share a repository or work must be dispatched, integrated, recovered, or retired.

Transform rough intent into an authoritative, testable specification and coherent dependency-aware task candidates. Use when clarifying requirements, defining non-goals and acceptance criteria, or preparing feature plans before task contracts.

Create or validate executable MAD task contracts with explicit authority, scope, dependencies, controlled artifacts, observable completion, checks, and handoff requirements. Use before any agent starts architecture, coding, QA, release, planning, or revalidation work.

Orchestrate a MAD work item through readiness, artifact-based dependency planning, execution waves, assignment, staleness, revalidation, QA, and release. Use when controlling Kanban flow or deciding what may run next or in parallel.