Everything this programme learned about writing and running quantum circuits with Quantinuum's Guppy language on the Selene emulator, compressed into one skill pack: v1 API rules, halfturn angle conventions, the noise models that actually exist, resumable per-row sweeps, unitary equivalence oracles, the offline TKET compile lane, and a gotcha list where every entry cost a debugging session.
It assumes Python 3.12+ and pip install "guppylang>=1.0" — Selene ships inside guppylang. The TKET lane needs pytket and pytket-quantinuum, runs fully offline, and never spends HQCs. The three downloads hold the same cards; only the manifest, install path and host notes differ.
Going online as well? The Nexus field guide covers login from a sandbox, the failures worth avoiding, cost guards and a costed dry run on H2-1LE.
Auditing an online run afterwards? The saved-Ref ledger shows every job this project has submitted and paid for, and which of them can still be re-attached to a sweep cell.
Hosted agent that builds and edits a web project in place.
Download · 133 KB.agents/skills/quantinuum/unzip quantinuum-lovable.zip -d .agents/skills/ # then: Settings → Skills → activate "quantinuum"
Carries the Lovable-specific cards: the .pydeps vendoring convention and why it must be gitignored, the Cloudflare Worker limits that force results to ship as committed JSON, and the gated-authoring protocol for turn rollbacks.
Verify — ask the agent to compile and run the 1-qubit smoke kernel at 64 shots. If it reaches for result() instead of output(), the skill is not loaded.
Self-hosted CLI/desktop agent from Nous Research, optionally on a local model.
Download · 124 KB~/.hermes/skills/quantinuum/ · or a skills.external_dirs entryunzip quantinuum-hermes.zip -d /tmp/qn cp -r /tmp/qn/quantinuum-hermes/skills/quantinuum ~/.hermes/skills/
Also ships the three research-team profiles — runner, analyst, scribe — as SOUL.md files, a mem0 MCP snippet for shared memory, and a daily frontier-digest cron prompt. One process per profile; hand-offs between roles are files, never chat summaries.
Verify — ask the agent to compile and run the 1-qubit smoke kernel at 64 shots. If it reaches for result() instead of output(), the skill is not loaded.
Resident gateway daemon with a fenced workspace and 50+ messaging channels.
Download · 134 KB<workspace>/skills/quantinuum/ · or ~/.openclaw/skills/ for a managed copyunzip quantinuum-openclaw.zip -d /tmp/qn openclaw skills install /tmp/qn/quantinuum # workspace openclaw skills install /tmp/qn/quantinuum --global # managed
The OpenClaw CLI does not install from zip or archive paths — unzip first and install the directory. Instructions load on demand: OpenClaw advertises the path and a sha256 marker, and the model reads SKILL.md when the frontmatter description matches the task.
Verify — ask the agent to compile and run the 1-qubit smoke kernel at 64 shots. If it reaches for result() instead of output(), the skill is not loaded.
A skill folder is executable trust: an agent will follow whatever it says. Read SKILL.md and the references/ cards before pointing an agent at them — that goes for this pack and every other one you install.
Four runnable notebooks — Guppy, Selene, tket and lambeq — assembled from the crawled upstream documentation rather than written from memory. Every code cell carries the page title and source URL it came from, each notebook closes with the pitfalls that apply to that library, and nothing in them touches credentials or spends a single HQC.
The language every kernel here is written in: qubits, arrays, structs, control flow, measurement and the v1 API.
Sentences to string diagrams to parameterised circuits, with lambeq's own training loop.
The emulator that runs those kernels: simulators, error models, seeds, logs and runtime debugging.
The compiler lane: circuit construction, compilation passes, routing and backend compile.
All four in one archive: quantinuum-notebooks.zip 15 KB. They are generated, not hand-written — a docs refresh re-emits them, and the diff shows which upstream example moved.
No agent, no network, no API key — just a model on your own machine. This is the whole programme flattened into one plain-text file: the 77 public pages with their descriptions, the 33 method cards, the design notes, and a digest of all 61 result files with their shot counts and column names. Load it once and the model can answer across the lot.
333 KB · roughly 85k tokens · built 2026-08-27. Fits whole into a 128k-context local model; chunk it for smaller ones.
Download · 333 KBcurl -O https://arunquantum.lovable.app/llms-full.txt ollama run llama3.1 "Answer only from this corpus. $(cat llms-full.txt)"
In LM Studio or Open WebUI, add it as a document to the chat instead of pasting it.
The short index in the llmstxt.org shape: one line per public page, for crawlers and for agents that would rather fetch pages than swallow the corpus.
View /llms.txtNote — both files are generated from the repository at build time, so the numbers in them are the same numbers on the pages. Nothing was written twice by hand.
One SKILL.md with the install steps and the numbered gotcha list, plus 33 reference cards loaded on demand so the agent only pays for what the task needs.
The skill is the working residue of the Nadarasa Reduction programme — a sequence of emulator gates (G1–G24) on Quantinuum's stack, and an NHS-facing arm applying the same instruments to discharge-flow and rota problems. Nothing in the pack is aspirational: each card describes something that was run.
Free to use and adapt, attribution appreciated. Guppy, Selene, TKET, H2 and Helios are Quantinuum products; this skill is an independent third-party pack and is not endorsed by Quantinuum.