Building with agents

mypackage exists so that summarising, smoothing and rescaling a numeric series is a call to a small, validated, pure function rather than a loop written again in every project. Coding agents tend to re-implement what they cannot see — a variance loop, a rolling median over a deque, a z-score scaler fitted on the wrong split — and each copy loses the validation, the boundary handling or the fit / apply discipline. The project ships three things that let them find the package instead.

The capability index¶

The capability index lists every public name in mypackage and mypackage.cli, grouped by the module that defines it, with a one-line summary. It is regenerated from the code (uv run poe capabilities) and checked in the test suite, so it never drifts.

The Claude Code plugin¶

The repository is a Claude Code plugin marketplace. In any project:

/plugin marketplace add jejjohnson/pypackage_template
/plugin install mypackage@mypackage

The plugin adds:

• series-analysis-with-mypackage (skill) — loads whenever a task summarises, smooths, rescales or cleans a numeric series: what lives where, the rules (inputs are never mutated, fit then apply, structural transforms, typed errors), a worked example, and a “don’t write it — use mypackage” table.

• mypackage-reuse-reviewer (subagent) — a read-only check of a diff for series code the package already provides, and for misuse.

llms.txt¶

For other agents and tools, the site serves llms.txt: a curated map of the package and its key pages.

Rules for your project’s AGENTS.md¶

Paste this into the agent instructions of a project that builds on mypackage:

## Numeric series: build on mypackage

This project uses mypackage for statistics, smoothing and transforms of
one-dimensional numeric series. Before writing a statistic, a rolling
window, a scaler or an input validator, search the capability index
(https://jejjohnson.github.io/pypackage_template/reference/capabilities/) or
`mypackage.__all__`, and compose what exists:

- Statistics: `mp.summarize` (mean, variance, std, median, extrema),
  `mp.quantile`, `mp.zscores`; streaming, `mp.RunningStats`.
- Smoothing: `mp.moving_average`, `mp.median_filter` (for spikes),
  `mp.exponential_moving_average`, with `padding=` from `mp.Padding`.
- Rescaling: fit `mp.Standardize` / `mp.MinMaxScale` on training data only
  and reuse the fitted object; chain steps with `mp.chain` / `mp.Pipeline`.
- Never mutate the caller's list; catch `mp.MypackageError`, not a bare
  `except`.

Working on the package itself¶

Contributors (and their agents) follow AGENTS.md in the repository: the module map, “reuse before you write”, the contracts and the tests that enforce them, and recipe skills for adding functions, transforms, CLI commands, guide pages and notebooks. Adopting the template for a project of your own is a recipe too: adopt-template in .claude/skills/.