A collection of Agent Skills that teach LLM agents
how to use hdlib — a Python library for
building Vector-Symbolic Architectures (a.k.a. Hyperdimensional Computing).
These skills make it easy for any skill-aware coding assistant (Claude Code,
GitHub Copilot CLI, and other tools that implement the open
Agent Skills standard) to write correct, idiomatic
hdlib code on demand — without burning context on the full library
documentation.
Target library:
cumbof/hdlibv2.1.0+
A skill is a directory containing a SKILL.md file with YAML frontmatter:
---
name: hdlib-vectors
description: Use when you need to create, combine, or compare hyperdimensional Vector objects with hdlib (binary/bipolar vectors, dist, normalize, dump). Covers hdlib.vector.Vector.
---
# Markdown body with detailed instructions and code examplesThe description is what the agent sees by default; it decides when to load
the skill. The body of SKILL.md is only injected into the agent's context
once the skill is invoked — so detailed reference material costs nothing
until it's needed.
hdlib-skills/
├── README.md # This file
├── CONTRIBUTING.md # How to add or modify skills
├── LICENSE # MIT
└── skills/ # All skills live here
├── README.md # Index with descriptions
├── hdlib-overview/ # Master skill - load this first
│ └── SKILL.md
├── hdlib-installation/
├── hdlib-vectors/
├── hdlib-space/
├── hdlib-arithmetic/
├── hdlib-distance/
├── hdlib-classification/
├── hdlib-feature-selection/
├── hdlib-hyperparameter-tuning/
├── hdlib-clustering/
├── hdlib-regression/
├── hdlib-graph/
├── hdlib-quantum-arithmetic/
├── hdlib-quantum-classification/
├── hdlib-quantum-advanced/
├── hdlib-analogical-reasoning/
├── hdlib-encoding-data/
├── hdlib-pitfalls/
└── hdlib-reproducibility/
Copy the skills/ directory under either of:
# Personal (available across all your projects)
cp -r skills/* ~/.claude/skills/
# Project (only this project)
mkdir -p .claude/skills
cp -r skills/* .claude/skills/After the copy, Claude Code picks up the skills automatically (no restart needed for personal/project additions).
Place the skills under your Copilot CLI skills path (consult the Copilot CLI
documentation for the exact location on your platform), or invoke them
manually by referencing the relevant SKILL.md.
Follow your tool's instructions for the open Agent Skills standard. The
directory layout (<skill-name>/SKILL.md with YAML frontmatter) is portable.
See skills/README.md for the complete index with
descriptions of every skill in this repository.
| Skill | Use when... |
|---|---|
hdlib-overview |
First touch with hdlib — gives the big picture and points to the other skills |
hdlib-installation |
Installing hdlib or wiring up its dependencies (numpy, scikit-learn, qiskit) |
hdlib-vectors |
Working with the Vector class (creation, operators, dist, normalize, dump) |
hdlib-space |
Organising vectors in a Space (insert, tags, links, search, persistence) |
hdlib-arithmetic |
Using the MAP operators: bundle, bind, subtraction, permute |
hdlib-distance |
Choosing between cosine, Euclidean, and Hamming distance for HD vectors |
| Skill | Use when... |
|---|---|
hdlib-classification |
Supervised classification with ClassificationModel |
hdlib-feature-selection |
HDC-based stepwise feature selection (forward/backward) |
hdlib-hyperparameter-tuning |
Parameter sweep on vector size and number of levels |
hdlib-clustering |
Unsupervised clustering with ClusteringModel (HDC k-means) |
hdlib-regression |
RegHD multi-model regression (RegressionEncoder, RegressionModel) |
hdlib-graph |
Encoding directed/undirected weighted graphs with GraphModel |
| Skill | Use when... |
|---|---|
hdlib-quantum-arithmetic |
Quantum MAP arithmetic (encode, bundle, bind, permute) and oracle compression |
hdlib-quantum-classification |
QuantumClassificationModel on simulators or IBM Quantum backends |
hdlib-quantum-advanced |
Compute-uncompute test, superposition_bundle, entangled_bind, Grover search |
| Skill | Use when... |
|---|---|
hdlib-analogical-reasoning |
Implementing role-filler analogies ("Dollar of Mexico"-style queries) |
hdlib-encoding-data |
Choosing how to encode tabular, sequence, or graph data into hypervectors |
hdlib-pitfalls |
Debugging unexpected results, type/size mismatches, normalisation issues |
hdlib-reproducibility |
Making hdlib runs deterministic with seeds |
Each skill was written against the actual source code of cumbof/hdlib
and the docstrings within it. The canonical code block from every skill
is exercised by test/smoke_test.py, which
passes against hdlib==2.1.0. See test/README.md
for how to run it locally. New skills must add a corresponding test
entry. See CONTRIBUTING.md for how to add or update a skill.
MIT — see LICENSE.