Architecture Skill
Harness Creator
Design safer, layered, production-ready agent harnesses instead of drifting into vague "AI agent" abstractions.
- Turns fuzzy runtime requests into clear module boundaries, implementation order, and safety rules.
- Pushes answers toward loop, tools, permissions, planning, memory, compaction, integrations, and verification.
- Useful for coding agents, CLI runtimes, MCP-enabled assistants, and subagent orchestration.
Observed effect
Without the skill, analysis tends to stall in clarification or broad architecture talk. With it, output becomes implementation-ready and source-aware.
Coordination Skill
Agent Teams Creator
Turn multi-agent swarm ideas into an explicit runtime with a task board, mailbox protocol, coordinator role, and verifier path.
- Separates task board as shared state from mailbox as transport.
- Makes spawn modes, teammate identity, coordination spine, and isolation stance explicit.
- Useful for coordinator-worker systems, verifier workflows, and protocol-driven team runtimes.
Observed effect
Without the skill, outputs often stop at file reading or generic swarm talk. With it, the protocol surface and team runtime model become concrete and buildable.
Design Skill
Stitch UI Bridge
Send frontend design prompts from Codex, Claude Code, or Cursor into Stitch and get raw exported code back through a local browser bridge.
- Reuses a logged-in Stitch browser session through CDP without exposing private profile data in the public repo.
- Tracks generation stages such as crafting, imagining, building the design system, and ready-to-export.
- Supports an explicit premium path for
Thinking with 3.1 Pro when the signed-in session has access.
Observed effect
Instead of vague "still waiting" automation, the bridge reports what Stitch is doing now and returns raw export code the caller can adapt downstream.
