Agent Skills that help developers using AI coding assistants build on Powabase — the multi-tenant AI Backend-as-a-Service. Agent Skills are folders of instructions and resources that an agent discovers and loads on demand to work more accurately and efficiently. Compatible with Claude Code and any agent that supports the Agent Skills open standard.
Powabase in one line: a single REST API that replaces the usual AI stack — a vector DB for RAG, an agent framework for tool use, a workflow engine for automation — alongside a full Supabase-style backend (Postgres + pgvector, PostgREST, Auth, Storage, Realtime). This skill teaches your assistant to drive all of it over plain HTTP.
SKILL.md is a cross-vendor open standard, so the same
skill installs across Anthropic, OpenAI, Google, and other coding agents.
| Tool | Install |
|---|---|
| Any agent (cross-agent installer) | npx skills add powabase-ai/agent-skills |
| GitHub Copilot / GitHub CLI | gh skill install powabase-ai/agent-skills |
| Google Gemini CLI | gemini extensions install https://github.com/powabase-ai/agent-skills |
| Claude Code (plugin) | claude plugin marketplace add powabase-ai/agent-skills then claude plugin install powabase@powabase-agent-skills |
| OpenAI Codex / Cursor / Windsurf / Cline / … | use a cross-agent installer above, or copy skills/powabase/ into the tool's skills directory |
The per-ecosystem manifests live in this repo: .claude-plugin/marketplace.json
(Claude), gemini-extension.json + GEMINI.md (Gemini CLI), and
skills/powabase/agents/openai.yaml (Codex) — all pointing at the single canonical
skill at skills/powabase/SKILL.md.
See PUBLISHING.md for where this is listed, how to submit it to each marketplace, and the roadmap for the consumer marketplaces (ChatGPT Apps, Gemini Enterprise) that run through the forthcoming Powabase MCP server.
powabase
End-to-end Powabase development skill covering the agentic /api/* surface and
the BaaS layer.
Use when:
- Building RAG: uploading documents (Sources), creating Knowledge Bases,
choosing an indexing strategy (
chunk_embed,full_document,page_index,graph_index,doc2json) and a retrieval method (vector, full-text, hybrid, tree search), reranking, and metadata enrichment. - Building agents: ReAct agents with builtin/custom/MCP tools, sessions, approval (human-in-the-loop) flows, and streaming.
- Multi-agent orchestration (supervisor / sequential / parallel coordinators).
- Workflow automation: block graphs, webhooks, scheduled triggers, the Copilot.
- Consuming Server-Sent Events for agents, orchestrations, and workflows.
- The Powabase BaaS layer: PostgREST, RLS, GoTrue auth, Storage, Realtime, and
direct Postgres — including the
ai.*schema. - Connecting and authenticating, choosing the right API key, handling billing / rate-limit / error responses, and knowing when to hand off to a human for Studio-only setup.
The skill ships a lean SKILL.md (connection, decision trees, the canonical RAG
flow, and the highest-value gotchas) plus detailed references/ files that the
agent loads only when relevant — keeping context small until depth is needed.
Content is synthesized from the Powabase docs. Because
the agentic /api/* surface is still evolving, the skill instructs agents to
verify shapes against the live docs rather than trusting a frozen snapshot.
skills/
└── powabase/
├── SKILL.md # required manifest (frontmatter) + core instructions
└── references/ # detailed, load-on-demand topic files
See CONTRIBUTING.md. In short: edit the Markdown, run
pnpm test, and open a PR with a conventional-commit title (feat: / fix:).
Each release packages every directory under skills/ into its own .tar.gz and
publishes a dist/index.json discovery document (agentskills schema v0.2.0)
alongside the tarballs, built by scripts/build-release.ts
and driven by Release Please.