madhav-anthropic

Madhav's Act and Transform skill, distributed as an installable Python package. It turns a task into a professional, evidence-driven prompt using five steps: professional standards, primary evidence, concrete methods, independent validation, and a copy-ready transformed prompt.

This is an independent instruction package, unaffiliated with Anthropic. It does not include a model, call an API, or require an API key. Python 3.10+ is required.

Install and use

Install the package, then run its command from your project folder:

python -m pip install --upgrade madhav-anthropic
madhav-anthropic

The second command creates skill.md directly in your current project folder. This default is available in version 0.1.1 and later. You install the package once per Python environment and run madhav-anthropic once in each project that needs the file. The file stays there between sessions.

pip install itself installs into the Python environment. Wheel installation does not provide a supported post-install hook for writing to the calling project directory. See the wheel installation specification. Assistant activation still depends on that assistant's skill configuration.

To display the instructions or export a skill folder for an assistant's loader:

madhav-anthropic show
madhav-anthropic export ./act-and-transform

The export command creates ./act-and-transform/SKILL.md as before.

Use the original trigger, act and transfrom, alongside your task, or explicitly ask to apply Act and Transform. Example:

act and transfrom: Help me design an evaluation for my image classifier.

Existing files are preserved. To deliberately replace the current project's file:

madhav-anthropic init --force

To replace an exported skill folder's file:

madhav-anthropic export ./act-and-transform --force

If the shell cannot find the command, use python -m madhav_anthropic instead:

python -m madhav_anthropic show
python -m madhav_anthropic
python -m madhav_anthropic export ./act-and-transform

Python API:

from madhav_anthropic import get_skill

instructions = get_skill()

Develop and build

skill.md at the project root is the authoritative source. The build copies it into the package as SKILL.md; there is no second source copy to maintain.

python -m venv .venv
.\.venv\Scripts\python.exe -m pip install --upgrade build twine
.\.venv\Scripts\python.exe -m build
.\.venv\Scripts\python.exe -m twine check --strict dist/*
.\.venv\Scripts\python.exe -m pip install --force-reinstall --no-deps .\dist\madhav_anthropic-0.1.1-py3-none-any.whl
.\.venv\Scripts\python.exe -m unittest discover -s tests -v

Publish to PyPI

1. Sign in at https://pypi.org/, verify your email, and enable two-factor authentication if needed.
2. Create an API token at https://pypi.org/manage/account/token/. For a first upload, use account scope; after creating the project, use a project-scoped token for subsequent releases.
3. From this folder, run:

.\publish.ps1

The script builds, checks, installs, and tests the release before uploading. At Twine's password prompt, paste your PyPI API token. The username is __token__. Enter the token only into that terminal prompt; do not put it in source files or chat. The script uses the real PyPI upload endpoint.

After a successful upload:

python -m pip install madhav-anthropic

PyPI names are shared globally. The requested name is not reserved until the first upload succeeds. If PyPI rejects it as unavailable, select another name and update the package metadata, version lookup, and documentation before building.

For another release, update version in pyproject.toml, then run publish.ps1 again. A published version cannot be overwritten with different files.

Packaging reference: https://packaging.python.org/en/latest/tutorials/packaging-projects/. Authentication reference: https://pypi.org/help/#apitoken.