Orchestrate Complex Objectives into Guided Sequences

Antigravity Workflows turns a complex objective into a guided sequence of skill invocations, one phase at a time. Instead of hand-picking each skill for a multi-phase goal, you name the outcome and Antigravity Workflows proposes the one or two best-matching workflows, then walks through each phase (for example plan, build, test, ship) while announcing the current step and its expected artifact. Before moving to the next step, the skill verifies completion criteria; if a check fails, it keeps the failure, fixes the relevant input or implementation, reruns the check, and only continues once it passes. If a prerequisite is unavailable, it reports the specific blocked step rather than silently working around it. It reads its workflow definitions from docs/users/workflows.md and data/workflows.json when those paths exist in the AAS repository, and falls back to a bundled set of workflow cards in a standalone installation - it does not invent missing files or fetch a silent replacement. At the end of a run it reports completed artifacts, validation evidence, and any remaining risks or next actions.

Use it when a request spans multiple skills that would otherwise need to be selected and sequenced by hand, or when the goal is explicitly multi-phase. Five default routes are built in: a product delivery request routes to ship-saas-mvp, a security review routes to security-audit-web-app, an agent or LLM product request routes to build-ai-agent-system, an end-to-end or browser testing request routes to qa-browser-automation, and a domain-driven design request routes to design-ddd-core-domain. Do NOT use it for a single, already-scoped skill invocation - it only asks clarifying questions when the choice materially changes scope, and it explicitly orchestrates rather than replaces the specialized skills it calls. It also depends on those referenced skills actually being available locally, and on environment access, credentials, or infrastructure it does not itself guarantee.

Input is the user's concrete outcome (for example "ship a SaaS MVP" or "audit application security"), which the skill maps to one of its default workflow routes or a bundled workflow card. Each bundled card defines its own inputs: shipping an MVP needs a target user, one primary journey, the current code, constraints, and a deployment destination; a security audit needs owned code or an explicitly authorized target and permitted test methods; building an agent system needs a task, tool permissions, data boundaries, and a measurable output; browser QA needs an existing route or journey, supported viewports, and the actual test runner; a DDD design pass needs business vocabulary, representative scenarios, and existing boundaries. Output is phase-by-phase: each step's artifact plus its verification, and a final summary of completed artifacts, validation evidence, and residual risks or blockers. For a reviewed skill-installation handoff, the output includes a concrete installer command run first with --dry-run against the exact IDs and release selected, and only repeated without that flag once the preview succeeds and the user has authorized it.

Antigravity Workflows is a meta-orchestrator: it calls other skills rather than doing the work itself, including brainstorming, concise-planning, workflow-automation, and verification-before-completion, plus the workflow-specific skills named in each default route - api-design-principles, frontend-design, e2e-testing-patterns, deployment-procedures, threat-modeling-expert, api-security-best-practices, frontend-security-coder, ai-agents-architect, llm-app-patterns, mcp-builder, agent-evaluation, vitest-skill, javascript-testing-patterns, browser-automation, fixing-accessibility, domain-driven-design, ddd-context-mapping, and ddd-tactical-patterns. Skill installation itself goes through a separate direct installer, invoked as an npm package:

npm exec --yes --ignore-scripts --package=agentic-awesome-skills@16.7.0 -- \
  agentic-awesome-skills --release 16.7.0 --path .agents/skills \
  --skills brainstorming,systematic-debugging --dry-run

Teams and individual developers who already use a library of narrow, single-purpose skills and want a guided default sequence for common multi-phase goals - shipping an MVP, running a security audit, building an agent system, doing browser QA, or designing a DDD core domain - without manually chaining each skill invocation by hand.