Getting started

The problem

Ask a hard question and most assistants pick the first reasonable-sounding answer and then defend it. That is pattern matching with confidence, not search.

What this skill changes

It holds several competing answers open at once, attacks each one with evidence or tools, and only settles when one is genuinely better supported. The branch bookkeeping is explicit rather than hidden, so a host can measure it.

It is a metaphor, not quantum computing. It runs on ordinary models and ordinary hardware; see the README.

Step 1 — verify what you installed

git clone https://github.com/Furox-Art/quantum-reasoning-skill.git
cd quantum-reasoning-skill
python docs/quickstart.py

Requires Python 3.10+. Exit code 0 means the skill contract parses and the shipped tooling runs. Full output and provenance: Quickstart.

If you installed from a package instead, verify the installed asset rather than the checkout:

pip install quantum-reasoning-skill
quantum-reasoning --validate      # Python console script

npm i quantum-reasoning-skill
npx quantum-reasoning --validate  # npm CLI

All three channels — npm, PyPI and a clone of main — serve 1.2.0. An earlier npm publish attempt failed with an E404 response to the PUT request (a publisher/credential configuration issue); the version was published after that was resolved.

--validate exits non-zero if the shipped SKILL.md is missing or malformed. The two CLIs do not share a flag set: --path prints the resolved asset path in the npm CLI only, and --skill-file PATH / --quiet exist in the Python one only. See the distribution section.

Step 2 — install SKILL.md in your host

SKILL.md is the deliverable. It is a Markdown file with YAML front matter; keep the front matter intact.

Your host supports Do this
A skill or plugin directory Copy SKILL.md into that directory as the skill entry point
Persistent system or developer instructions Load the whole file into that instruction layer before the task (example)
An explicit agent loop you control Map observable branch measurements onto the reference controller (example)

Capability requirements in full: Compatibility.

To check a copy landed correctly, keep the front matter and verify the two required fields:

python -c "import pathlib;t=pathlib.Path('SKILL.md').read_text(encoding='utf-8');print('name: quantum-reasoning' in t);print('chars:',len(t))"

Step 3 — ask a question worth branching over

Should we migrate our monolith to microservices? Keep materially different migration
strategies open, say what evidence would refute each one, use whatever tools are
available to test them, and give me only the best-supported recommendation plus the
uncertainty that remains.

Good prompts name the branching behaviour you want, because not every host will infer it. More shapes: usage examples.

What you get

  • Several genuinely different approaches explored, not paraphrases
  • Each judged on evidence rather than plausibility
  • Contradicted paths pruned; recoverable ones kept dormant and revived on new evidence
  • A conclusion with the remaining uncertainty stated, and alternatives named if they genuinely survive

When not to use it

Simple questions. If the answer is obvious, branching just burns tokens and latency. SKILL.md scopes this: use it for difficult, ambiguous, multi-step or high-stakes problems.

What you do not get

A measured improvement. No baseline-versus-skill evaluation has been published for this repository, and the reference thresholds are uncalibrated defaults. If you run the evaluation yourself, CONTRIBUTING.md lists the provenance every contributed number must carry.