Brave Bot documentation

The documentation site for Brave Bot, a general-purpose agent with structural resistance to indirect prompt injection.

Read it at brave-experiments.github.io/bravebot-docs.

Built with Docusaurus and published from main on every push.

make init     # install dependencies and link agents/ into .claude/ and .bravebot/
make start    # serve locally with live reload

make on its own lists every target. The npm scripts still work if you prefer them: npm install, npm start, npm run build.

make build

Generates static content into build/, which can be served by any static host. Broken links and anchors are build errors rather than warnings, so a clean build is also the correctness check, and it is the whole of what CI runs.

Everything here describes behaviour that is specified clause by clause in the mini-specs in the main repository. Where the two disagree, the specs are the source of truth: fix this site rather than documenting around it.

docs-updated-to-sha records the bravebot commit this site was last brought up to.

make docs-updated-to-sha   # where the docs stand against bravebot
make docs-changes          # what has landed since

Both read a bravebot checkout at ../bravebot, or wherever BRAVEBOT_REPO points. To fold the gap in, run the update-docs skill: it reviews the intervening commits, updates the pages that went stale, then records the new commit and commits that too.

agents/ is the checked-in source of truth for skills, AGENTS.md and each tool's settings. No tool reads it directly. make init symlinks it into .claude/ for Claude Code and .bravebot/ for bravebot, so a skill is written once and both find it. The links are generated and gitignored; make agents shows their state and make unlink removes them.

MPL-2.0