Brain template (for sessions working on the template)

No local/ folder here? Then this is a copy of the template that someone wants as their own brain, not a session developing the template: run brain-setup and ignore the rest of this file. Speak plainly; the person may not know git or a terminal (docs/guides/start-here.md).

This folder is the template, not a brain: brain.json has role: template and prefix template. README.md says what the template is; this file says how to work on it.

• brain-* skills are the product. They ship to every brain, and are edited here, directly, through brain-write-skill. This file has no managed block, and the template's .claude/settings.json wires the skill gate, so a write to a skill file is refused until brain-write-skill has run. Edit skills with brain-write-skill, scripts with brain-write-script.
• template-* skills never ship. template-make-brain and template-intake work on the template and stay in it. A new skill asks first: does it ship (brain-*) or stay (template-*)?
• Shipped means copied into brains: _meta/** and .claude/skills/brain-*/**, plus the managed blocks and the skeleton in _meta/skeleton/. Everything else here is template-only: this file, README.md, brain.json, tests/, docs/, local/, and the public repo's LICENSE, CONTRIBUTING.md, SECURITY.md, CODE_OF_CONDUCT.md and .github/ (init removes them all from a copy; _meta/LICENSE carries the licence into every brain). Skeleton files that are dotfiles or sit under .claude/ are stored renamed (gitignore, env.example, claude-settings.json, mcp.json).
• Run the test suite before committing: PYTHONDONTWRITEBYTECODE=1 python3 -m unittest discover -s tests -t . -v. Then _meta/brain doctor (it needs local/ and the pre-commit hook: _meta/brain hook install).
• Nothing specific to where the template came from goes in a tracked file: no organisation, product, system or person from the brains it was distilled from, and no absolute home path. tests/test_origin.py scans every tracked file against local/origin-terms.txt (ignored, so the terms are never written down in the repo). Write for any brain.
• local/ is never committed. It holds notes, origin-terms.txt and brains.txt (the brains made from this checkout, one absolute path per line, which template-intake reads).
• Brains get changes through update, which reads the latest commit and not the working tree. A change brains should receive gets a _meta/CHANGELOG.md entry; one that answers an upstream suggestion cites its id (- landed: <id> — <summary> or - declined: <id> — <reason>).
• Before publishing the template, set template.source in brain.json to the published URL, so init in a clone of it starts a fresh history and never keeps the template as origin.
• Commit through brain-sync. Subjects are meta(<area>), skill(<name>), docs(<area>) or test(<area>). Never push unless asked.