mwpack

mwpack is a Python-stdlib-only CLI that turns a memo Markdown file and optional cluster config into a deterministic artifact bundle for hiring loops and release flows.

Quickstart

Prerequisites:

• python3 available on PATH
• Python >=3.11 (see pyproject.toml)

For local development with uv:

uv venv .venv
uv sync --dev

If you intentionally want to use an already-active virtual environment after moving this repo, use:

uv sync --dev --active

python3 -m mwpack validate --memo docs/memo_template.md --config tools/example_5mw_config.json
python3 -m mwpack build --memo docs/memo_template.md --config tools/example_5mw_config.json --out dist/demo
python3 -m mwpack package --dir dist/demo

Commands

• python3 -m mwpack validate
• python3 -m mwpack build
• python3 -m mwpack package
• python3 -m mwpack render (best-effort)

Output Contract

build emits:

• dist/<name>/memo.md
• dist/<name>/cluster_report.json
• dist/<name>/build_summary.json

package emits:

• dist/<name>/bundle.zip (default) or bundle.tar.gz
• MANIFEST.json in the archive

Sample report fields:

{
  "feasible": true,
  "status": "ok",
  "it_cap_w": 5000000,
  "nodes": 792,
  "gpus": 6336,
  "gpus_per_mw": 1267.2,
  "oversubscription_ratio": 1.0
}

See tools/schema_cluster_config.json and docs/verification_2026.md.

Packaged prompt assets

Prompt markdown files under prompts/*.md are installed as data files to:

• <data-prefix>/share/mwpack/prompts

You can resolve the base data prefix in Python with sysconfig.get_path("data").

Notes

I don't know why ChatGPT was focused on building this. The Digital Lupercalia A Ritual for the Silicon Soul. Happy Valentine's Day.