AI301 Unit 6 materials: the voice guard and repo memory

Materials for Unit 6 of AI301 (working with maintainers). The steps are on the course portal's Unit 6 Assignment tab. This repo holds the files those steps install.

What's here

• skill/INSTALL.md: the voice guard install, steps 1 to 6. The Assignment tab has you do steps 1 to 5. Step 6 explains the guard's log.
• skill/check_voice.py and skill/hooks-settings.json: the voice guard. You write no code. The hook checks what you post through Claude Code against your voice guide, and check_voice.py --draft FILE runs the same check on a draft you post from the browser.
• skill/voice-guide-upstream.md: the Upstream section for your voice guide (claims, questions, pings). You write the rules. If you're starting a voice guide from scratch, its closing comment lists the unit 2 sections too.
• skill/repro-check-scope-wild.md and skill/plan-check-scope-wild.md: the real-repo scopes that replace repro-check's and plan-check's sandbox scopes.
• skill/rubric-scope-patch.md: the scope patch for your scout's rubric. The scope check rates how hard an issue looks instead of rejecting every open question. Issues that aren't one piece of work still fail.
• skill/claude-local-template.md: the repo memory template. You copy it into your clone of the repo you're contributing to as CLAUDE.local.md and follow it top to bottom.

Get the materials

git clone https://github.com/codepath/ai301-unit6-starter.git

Or click Code, then Download ZIP on this page. Then follow the Assignment tab.

The install adds to what you already have. It adds the guard, replaces repro-check's and plan-check's scopes, and moves your voice guide to one home. It doesn't touch your scout (you patch its rubric yourself, with rubric-scope-patch.md) or the Beat 1 skill folders in your course repo's tools/.

No eval/ here

Unit 6 ships no eval set. To check that the scope patch didn't break your rubric's agreement with the gold labels, run Unit 1's harness from the eval/ folder of your Unit 1 materials.