banthis

banthis

Persist "banned behaviors" into CLAUDE.md / AGENTS.md so AI coding agents remember never to repeat annoying patterns across sessions.

When an agent does something frustrating repeatedly (hedging language, over-explaining, editing migration files directly, etc.), banthis turns that frustration into a permanent, high-priority rule that wins over future user requests.

Installation

npm install -g github:agent-sh/banthis

Or use via npx:

npx --yes github:agent-sh/banthis add "No hedging" "Do not start responses with 'To be honest' or 'I think' — it undermines confidence."

Core Commands

• banthis <title> <rule> — Quick add (shortcut for add)
• banthis add <title> <rule> — Add or update a banned behavior
• banthis list — List current bans
• banthis remove <title> — Remove a ban
• banthis init — Install the meta-rule so agents know to invoke banthis automatically
• banthis install-command — Drop the /banthis slash command into .claude/commands/

Flags

• -g, --global — Target ~/.claude/CLAUDE.md (user-wide)
• --file <NAME> — Target specific filename
• --dir <PATH> — Target specific directory

How It Works

banthis maintains a managed section in CLAUDE.md or AGENTS.md (between <!-- banthis:start --> and <!-- banthis:end -->).

Rules added here are treated as hard prohibitions — the agent is instructed that these rules have higher priority than the current user request.

This pairs well with positive knowledge tools such as axiom and with agnix for validation.

Philosophy

Negative rules ("never do X") are some of the highest-leverage instructions you can give an agent. They are cheap to write, extremely reliable, and survive long context windows better than complex positive instructions.

banthis makes it trivial to capture these in the moment of frustration so they persist forever.

Related Projects

• system-prompt-curator — For high-quality system prompts
• skill-curator — For high-quality SKILL.md files
• agnix — Linter and validator for agent configurations
• axiom — Positive institutional memory and decision capture

License

MIT