Bully

CLAUDE.md asks. Bully enforces.

When you tell Claude Code “don’t use console.log” or “no as any casts” in CLAUDE.md, it forgets. Not on the next edit, but eventually. The instructions are advisory; nothing checks the diff.

Bully wires a PostToolUse hook into Claude Code. Every Edit and Write runs through a pipeline that checks the change against .bully.yml. Error-severity rules block the tool call by exiting with code 2, so Claude has to fix the violation before the edit lands. Warnings surface in the transcript without blocking.

Three engines, one config

Rules in .bully.yml pick an engine based on what they need to check:

• script runs a shell command (grep, jq, a project linter). Deterministic; exit code wins.
• ast runs a structural pattern through ast-grep, so it ignores formatting variants, comments, and string contents.
• semantic ships the unified diff plus a plain-English description to a subagent (bully-evaluator) that returns a structured verdict.

The first two cover the rules a regex can express. The third covers the ones it can’t, like “don’t derive state from props with useEffect” or “inline single-use variables.” All three share one trigger, one output format, and one fix loop, across every language in the repo.

What a config looks like

schema_version: 1

rules:
  no-console-log:
    description: 'No console.log in committed source. Use the project logger.'
    engine: script
    scope: ['src/**/*.ts', 'src/**/*.tsx']
    severity: error
    script: "grep -nE 'console\\.log\\(' {file} && exit 1 || exit 0"

  no-any-cast:
    description: 'No `as any` casts. Use a precise type or `unknown` plus narrowing.'
    engine: ast
    scope: ['src/**/*.ts', 'src/**/*.tsx']
    severity: error
    pattern: '$EXPR as any'

  prefer-derived-state:
    description: >
      React components should not use useEffect to derive state from
      props. Compute the value directly during render (or with useMemo
      if expensive). Effect-based derivation causes unnecessary renders
      and stale reads.
    engine: semantic
    scope: 'src/**/*.tsx'
    severity: warning

Adopting bully in an existing repo with pre-existing violations would normally light up every old problem. bully baseline-init records the current state in .bully/baseline.json so only new violations block edits.

Install

Bully ships as a Claude Code plugin:

/plugin marketplace add https://github.com/dynamik-dev/bully
/plugin install bully

That installs the skills (bully-init to bootstrap a config, bully-author to add or modify rules, bully-review to surface noisy or dead rules from telemetry), the evaluator subagent, and the PostToolUse hook. Restart Claude Code, then run bully doctor to verify.

GitHub