Changelog

0.2.0 (2026-10-02)

The kit now covers the points where an agent's work reaches a person.

New skills:

• push-check: scans unpushed commits for secrets, malicious code, personal information and AI credit lines, blocks the push on a serious finding, and installs as a pre-push hook.
• proof-before-done: reports work as done only with a tool result, and outside facts only with a source.
• deliverable: drafts a graded or shared document from its rubric, at an agreed length.
• handoff: writes a handoff note when a session or tool ends, and picks it up in the next one.

Changed skills:

• quiz: drafts for approval go in full in the reply text, and reversible steps go ahead without a question.
• variant-lab: judges motion in a headless or foreground browser, and lays out media variants on one comparison sheet.
• writing-check: points graded documents to deliverable, and has a third eval case.

Scripts, CI and install:

• scripts/validate.py checks every skill folder, eval file, README row and plugin manifest.
• docs/evals.md explains how to run old-vs-new eval runs with any agent.
• CI runs the tests and validate.py on Python 3.10 and the latest 3.x.
• Manifests for GitHub Copilot CLI, Qwen Code, Gemini CLI, Codex and Cursor, with install steps and check dates in the README.

0.1.0 (2026-09-27)

First release: quiz, variant-lab and writing-check.