Agent skills for building quantum computing workflows. Equip your AI coding agent to design circuits, run simulators, orchestrate quantum optimization, and pair-program inside live aqora-hosted marimo notebooks. Works with Claude Code, opencode, pi, GitHub Copilot, Cursor, Codex, and any agent that reads the Agent Skills open standard.
The repo treats the platform (aqora workspaces) and the domain (quantum computing) as two complementary axes. Platform skills know how to drive a live workspace. Domain skills know qiskit, pennylane, circuit optimization, and simulator choice. Together they cover the full loop from "sketch a circuit" to "run it on a simulator in an aqora workspace and compare results."
| Skill | Purpose |
|---|---|
aqora-workspace |
Pair-program inside a live aqora-hosted marimo workspace. List workspaces, execute code in the remote kernel, create and edit cells. The entry point for running quantum code on aqora. |
quantum: qiskit and pennylane authoring, circuit construction and transformation, simulator selection (numpy, qiskit, pennylane), gate-count and depth optimization, QAOA and VQE templates. Pairs withaqora-workspacewhen the agent needs to run circuits on aqora infrastructure.- Additional platform skills as the aqora API grows: dataset management, competition submission, and workspace search.
- Additional domain skills for adjacent quantum tooling as the ecosystem evolves.
/plugin marketplace add aqora-io/skills
/plugin install aqora-skills@aqora-io-skills
Enable auto-update to stay current:
/plugin -> Marketplaces -> aqora-io-skills -> Enable auto-update
Works with any agent that supports the Agent Skills open standard, including opencode, pi, GitHub Copilot, Cursor, and Codex.
Install everything:
npx skills add aqora-io/skills
Install a single skill:
npx skills add aqora-io/skills --skill aqora-workspace
If you do not have npx but have uv:
uvx deno -A npm:skills add aqora-io/skills
Clone this repository and symlink individual skill directories into your agent's skills path. Common locations:
| Agent | Path |
|---|---|
| Claude Code | ~/.claude/skills/<skill-name> |
| opencode | ~/.config/opencode/skills/<skill-name> |
| pi | ~/.pi/agent/skills/<skill-name> |
| GitHub Copilot | ~/.copilot/skills/<skill-name> |
Only the leaf skill directory matters. The skills/ parent is not copied.
After installing, log in to aqora so the skills can authenticate:
pip install aqora-cli
aqora login
Then ask your agent a workspace question, for example:
- "list my aqora workspaces"
- "in my portfolio-optimizer workspace, add a cell that runs QAOA on the pennylane simulator and plots the efficient frontier"
- "open the numpy-simulator workspace and sanity-check the Hadamard gate"
The agent loads the relevant skill, targets the right workspace, and starts working. Cells appear live in the workspace UI.
Skills in this repo follow three rules.
- Portable first, Claude Code second. Skills are authored to the Agent Skills open standard. Claude Code manifests sit on top as an overlay. If a piece of guidance only applies to one agent, it belongs in a reference file, not the skill body.
- Progressive disclosure. The body of
SKILL.mdis short and action-oriented. Depth lives inreferences/*.mdfiles the body links to. Only the body enters context on invocation. References load on demand. - Scripts own the machine, kernels own the state. Bundled scripts run on the user's machine and talk HTTP. Business logic lives in the remote kernel, driven through
marimo._code_mode.
See CONTRIBUTING.md for authoring conventions, frontmatter rules, and the validation script CI runs on every pull request.
The execute-code script pattern is adapted from marimo-team/marimo-pair. Credit to the marimo team for the design.
MIT. See LICENSE.