Steerhook — Development skill for Claude Code
🛞 Rules that block or warn before Claude Code runs a tool and tell Claude the alternative.
How to install Steerhook
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open meganemura/steerhook and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Steerhook does
🛞 Rules that block or warn before Claude Code runs a tool and tell Claude the alternative. A fork of Anthropic's hookify plugin: rule messages reach Claude, and rules live in ~/.claude/steerhook/ for every project.
Alternatives in Development
- Hookify — Rule-based hook system for Claude Code 14k ★
- Open Agent SDK TypeScript — Agent-SDK without CLI dependencies, as an alternative to claude-agent-sdk, completely open source 2.6k ★
- Dario — A local OpenAI- and Anthropic-compatible proxy that brings your Claude Pro/Max subscription to Cursor, Cline 485 ★
README
steerhook
[日本語](README.ja.md)
A hook that steers. Write a rule that names a command pattern you do not want Claude to run and the form you want instead. Before Claude runs a tool, steerhook checks the call against your rules. A rule can block the call or let it through with a warning. In both cases the rule's message reaches Claude, so Claude learns the alternative at the moment it matters.
Rules are yours. They live in `~/.claude/steerhook/` and apply in every project. A project's own rules are never read.
steerhook is a fork of Anthropic's [hookify](https://github.com/anthropics/claude-plugins-official/tree/main/plugins/hookify) plugin for Claude Code (Apache 2.0, see `NOTICE`).
What is different from hookify
- The rule message reaches Claude. hookify sends the message only to the
user. Claude sees "denied" and cannot learn the alternative the rule asks for.
steerhook sends the message as
permissionDecisionReason(block) oradditionalContext(warn). The user still sees it assystemMessage. - Rules live in
~/.claude/steerhook/*.md. hookify reads rules only from the project. Rules about your own tools apply in every project, so they need one place. The/steerhook:addcommand and the writing-rules skill write there. SetSTEERHOOK_RULES_DIRto use a different directory. - A project's own rules are never read. hookify lets a project override a user rule by name. steerhook does not: opening a project is not the same as trusting it, and a rule file there could replace or switch off a user's rule with no confirmation.
- The PostToolUse hook is removed. Rules read only the tool input, so the PostToolUse hook repeated the PreToolUse check and put the message in the transcript twice.
- A bash rule matches what the shell runs. hookify matches the raw
command text, so
codex execinside a quoted argument and awhileloop inside a heredoc body both fire the rule, although the shell runs neither. steerhook
Related Skills
Clear Init
Install CLEAR-100 into the current repo or agent. Writes the rule block into AGENTS.md, CLAUDE.md, Cursor rule
Cc Agents Kit
Claude Code plugin marketplace. guard-hooks: five defensive hooks that block dangerous shell commands and secr
Hookify Configure
Enable or disable hookify rules interactively
PM Insight
Insight synthesis — unifica ricerca quant+qual+macro negli 8 building block strategici (Job, Alternative, Cate
Sensitive Canary
Claude Code hooks that guard secrets and PII before they reach the Anthropic API
Synthetic Customer Discovery
A Claude skill for honest synthetic customer discovery: grounded personas, anti-collapse interview rules, and
Related Agents
Opengrep Rule Engineer
Expert in OpenGrep (FOSS Semgrep fork) and Semgrep rule authoring. Writes precise patterns, taint-tracking rul
Issue Reviewer
READ-ONLY checklist reviewer for the craft-issue pipeline — dispatched at Stage 6's Checklist Review Gate imme
Rule Advisor
Selects optimal rulesets for tasks and performs metacognitive analysis. Use PROACTIVELY before implementation