shipwright-kit 0.6.0

Shipwright — AI-agent dev framework + import-light design/eval/security library

Shipwright

Shipwright is two things that share one repo:

• an installable, import-light Python library of shared dev-tooling runtime — shipwright_kit.design (severity tiers + accessible output), shipwright_kit.eval (detection-quality eval harness), shipwright_kit.security (a security pack);
• an AI-agent-operated development framework — reusable CI/CD, a Copier scaffolder, quality gates, and bundled agent skills + personas — that dogfoods the library and the gates it hands to the projects built with it.

The library is consumed today by two real tools: barb and sift both import shipwright_kit.eval to run their detection-quality gates.

Install

The library is not on PyPI — the bare name shipwright belongs to an unrelated project, so the published distribution is shipwright-kit and the import name is shipwright_kit. For now, install from git:

uv pip install "git+https://github.com/duathron/shipwright@main"
# then: import shipwright_kit

[!NOTE] Pin a release tag instead of @main for reproducible builds once a tagged release of the shipwright-kit distribution is cut. Do not pip install shipwright from PyPI — that is a different, unrelated package.

The security pack needs no extra — it ships with the base install and registers through the shipwright_kit.packs entry point.

Library quickstart

Run an eval gate — score a classifier against a labeled corpus and fail if it misses a floor (the exact pattern barb and sift use):

from shipwright_kit.eval import Sample, evaluate, gate

corpus = [Sample("phish-login", "phishing"),
          Sample("example.com", "benign"),
          Sample("secure-phish", "phishing")]

result = evaluate(
    lambda text: "phishing" if "phish" in text else "benign",
    corpus,
    positive_pred=lambda pred: pred == "phishing",
    positive_expected=lambda label: label == "phishing",
)
print(result.precision, result.recall)        # 1.0 1.0
gate(result, min_precision=1.0, min_recall=0.9)  # raises EvalGateError if below

Use the shared severity tiers — one generic scale tools map their own verdicts onto, with accessible (Unicode-or-ASCII) labels:

from shipwright_kit.design import Severity, tier_label

Severity.OK, Severity.INFO, Severity.NOTICE, Severity.WARN, Severity.CRITICAL  # IntEnum 0..4
print(tier_label(Severity.CRITICAL))  # ✗ CRITICAL
print(tier_label(Severity.OK))        # ✓ OK

import shipwright_kit pulls in no rich or pyfiglet — the heavy deps load lazily only when you actually render. Full API: docs/library.md.

The framework

The repo contains no project code. Projects stay their own Git repositories and their own packages; locally you clone each into the gitignored projects/ directory, where a uv workspace ties them together for development:

shipwright/
├─ shipwright_kit/          # the importable library (design / eval / security)
├─ tooling/ruff-base.toml   # single source of truth for lint rules
├─ templates/               # Copier scaffolder (python-cli) + release config
├─ skills/ · personas/      # the agent operating layer (scaffold, onboard, review …)
├─ .github/workflows/       # reusable python-ci.yml + python-release.yml (SHA-pinned)
└─ projects/                # GITIGNORED, local-only — your projects plug in here

Work is promoted through gates; failing a rung blocks promotion:

commit → lint + unit (auto) → build → dogfood + eval (auto) →
QM gate (manual) → beta sign-off (manual) → release

The reusable CI/CD that wires these gates, the Copier scaffolder (templates/), and the agent skills (skills/) and personas (personas/) all ship now. This repo runs the exact gates it gives the projects built with it.

Framework quickstart

Requires Python 3.11+, uv, and just.

uv sync --dev              # create the dev environment
uv run pre-commit install  # install the local gate
just lint                  # ruff check + format-check
just test                  # pytest

The local pre-commit gate runs the same lint/format/secret checks as CI; add just test (and uv build) for the test and build rungs CI also enforces.

Docs

• docs/library.md — per-module API reference (design / eval / security)
• docs/release-policy.md — SemVer + release policy
• docs/ci-cd.md — the reusable CI/CD workflows
• CHANGELOG.md

Security

Report vulnerabilities privately via GitHub's private vulnerability reporting (repo Security tab → Report a vulnerability). See SECURITY.md.

License

MIT © 2026 Christian Huhn