Hookify for Claude Code Review 2026: Pros, Cons, Install

What you get in one minute#

Hookify is a free Anthropic plugin that turns a plain-language request ("warn me when you run rm -rf") into a small rule file that blocks or flags Claude's commands and edits, but in its current code several rule types do not work and it can fail silently.

Logo: the official Claude Code logo from Anthropic's press kit (anthropic.com/press-kit)

The marketplace page shows 60,376 installs (October 2026). We found 103 GitHub issues with "hookify" in the title, 76 of them open.

Verdict	Detail
Useful?	Niche. Good idea, handy for simple command and edit guards, but unreliable on Windows and for prompt and stop rules.
Best for	Mac or Linux users who want a few regex guards on Bash commands and file edits without writing hook JSON.
Skip it if	You are on Windows without `python3`, want a hard security barrier, or need prompt or stop rules.
Made by	Anthropic, in its official marketplace. The README says MIT, the LICENSE file says Apache 2.0.
Version + date	No version number in `plugin.json` or the marketplace entry. Last change to the folder: 20 May 2026 (commit b58bdbf). Read at repository head 41f619d, 27 October 2026.
Install command	`/plugin install hookify@claude-plugins-official`
Token cost	No published figure; run `claude plugin details hookify`. Our rough count: a few hundred tokens at start (one skill, four commands, one agent description). It also starts a Python process on every tool call.

What it does#

You describe what Claude should not do, Claude writes a rule file, and from then on Python hooks check every action against it.

/hookify <what to stop> writes .claude/hookify.<name>.local.md in your project. Without text, it reads your recent messages for corrections and proposes rules.
Rules apply on the next tool call. A rule has an event (bash, file, stop, prompt, all), a regex pattern or a list of conditions, and an action: warn (default) or block.
Adds 4 commands (/hookify, /hookify:list, /hookify:configure, /hookify:help), 1 skill (writing-rules), 1 agent (conversation-analyzer) and 4 hooks: PreToolUse, PostToolUse, Stop and UserPromptSubmit.

How it works in the background#

The plugin only writes markdown; four tiny Python hooks do the checking every time Claude Code reaches an event. The picture follows one rule.

How hookify works: you run the hookify command, Claude writes a rule file into your project, later a tool call, a stop or a prompt starts a Python hook, which reads the rule files, tests the regex and then does nothing, warns on screen or blocks the action

No matcher in hooks.json: the hooks run on every tool call, even Read.
Warn versus block: a block returns permissionDecision: deny. A warning returns systemMessage, which Claude Code's hooks reference describes as shown to the user.

Under the hood: dissection#

Hookify is 64% Markdown prompts and 34% Python (about 850 lines), with no dependencies beyond the Python standard library.

The file tree#

Codetext

hookify/
├── .claude-plugin/plugin.json     name, description, author (no version)
├── hooks/
│   ├── hooks.json                 4 hooks, 10 s timeout, no matcher
│   ├── pretooluse.py              Bash/Edit/Write/MultiEdit events
│   ├── posttooluse.py             same checks after the tool ran
│   ├── stop.py                    stop rules
│   └── userpromptsubmit.py        prompt rules
├── core/
│   ├── config_loader.py           finds and parses the rule files
│   └── rule_engine.py             matches rules, builds the JSON answer
├── commands/                      hookify.md, list.md, configure.md, help.md
├── skills/writing-rules/SKILL.md  rule syntax, loaded by every command
├── agents/conversation-analyzer.md
├── examples/                      4 sample rules
├── README.md, LICENSE, .gitignore
└── matchers/, utils/              empty packages (only __init__.py)

File	Language	What it does	When it runs
`hooks/*.py`	Python	Read hook JSON from stdin, load rules, print a JSON answer, always exit 0	Each event
`core/config_loader.py`	Python	Home-made YAML reader for the rule files	Each event
`core/rule_engine.py`	Python	Regex and operators; block beats warn	Each event
`commands/*.md`	Markdown	Prompts that tell Claude how to create, list or toggle rules	When you type the command
`skills/writing-rules/SKILL.md`	Markdown	Rule format reference	Loaded by the commands
`agents/conversation-analyzer.md`	Markdown	Finds your corrections in the chat	`/hookify` without text

A bar chart of hookify's size in lines: command prompts 616, writing-rules skill 374, README 340, conversation-analyzer agent 185, rule engine 313, rule loader 297, hook scripts 237, example rules 68 and JSON files 57

Chart: line counts of each file at commit 41f619d, read 27 October 2026. Teal is Markdown, orange is Python.

Two key excerpts#

hooks/hooks.json, one of four similar entries:

File contentjson

"PreToolUse": [{ "hooks": [{ "type": "command",
  "command": "python3 \"${CLAUDE_PLUGIN_ROOT}/hooks/pretooluse.py\"", "timeout": 10 }] }]

core/rule_engine.py, the block answer:

Codepython

"hookSpecificOutput": {"hookEventName": hook_event, "permissionDecision": "deny"},
"systemMessage": combined_message

Disk and network#

Reads: .claude/hookify.*.local.md relative to the folder the hook starts in, and for stop rules the whole transcript file.
Writes: nothing from the hooks; Claude writes the rule files.
Network: none. Needs: python3 on the PATH. The README says Python 3.7+, but the code needs 3.9 or newer (an open issue reports a crash on 3.8).

What we ran ourselves#

We ran the hook scripts on a Windows PC in a throwaway folder with sample rule files (Python 3.13, run from a shell, not inside Claude Code).

Test	Result
`bash` rule `rm\s+-rf`, action block, command `rm -rf /tmp/x`	Denied, but no reason sent to Claude
Same rule, command `rm -r -f /tmp/x`	Not caught (regex only sees the text)
`file` rule `console\.log\(` on an `Edit`	Warning returned
Same rule on a `Write` of a new file	No match
`prompt` rule with the README's `pattern:` form	No match
Run from a sub-folder with no `.claude/`	No rules found
Start-up cost, 5 runs with no rules	About 0.27 seconds each

What users say#

We read the first part of 27 reports in detail (24 complaints, 1 mixed, 2 praise) and sorted the titles of all 103 hookify issues into themes. Reddit was not read. The split is skewed: bug trackers hold bug reports, and many read like code audits.

A horizontal bar chart of the 103 hookify GitHub issues by theme: 20 rules fire on the wrong event or never fire, 20 Windows encoding and path problems, 20 install and import errors, 12 rule files not found, 11 warning or block text does not reach Claude, 11 feature requests and docs, 9 rule engine quirks

Chart: issues with "hookify" in the title in anthropics/claude-plugins-official and anthropics/claude-code, themes labelled by DIY Lab Hub, 27 October 2026. Many install errors date from 2025.

Voice	What they say
Praise	A Hacker News commenter (February 2026) and another comment (February 2026) recommend it for guardrails and blocking commits.
Warnings never reach Claude	Issue 20747 (January 2026, 6 thumbs up) and issue 5290 (August 2026).
Block gives no reason	Issue 6314 (September 2026): Claude only sees a generic "Blocked by hook".
Prompt rules dead	Issue 2102 (May 2026): the engine reads `user_prompt`, the docs say `prompt`.
Rules silently not loaded	Issue 6339 (September 2026): search starts in the live folder, not the project root.
Windows	Issue 85 (December 2025, 5 thumbs up): `python3` not found; issue 6092 (September 2026): non-ASCII rule files fail.

Pros: why it is useful#

It makes the most common guardrail (never do X) a one-sentence job and keeps the rule in a file you can read.

No JSON, no code: you can write or edit a rule in a minute; rules apply immediately.
Learns from corrections: /hookify proposes rules from the chat.
Transparent: about 850 lines of Python, no network.

Cons: why it is not useful#

The idea is better than the current code, and a rule that silently does nothing is worse than no rule.

Silent gaps: we reproduced Write ignored by simple file rules, and prompt rules do not fire.
Not a barrier: regex on text misses rm -r -f, /bin/rm and scripts. Claude Code's own Bash(rm *) deny rule also matches inside compound commands and after variable assignments (permissions docs).
Claude may never read your warning, and a block carries no reason.
Windows is a weak spot: python3, encoding and backslash-path reports.
Cost: one Python start per tool call (about 0.27 seconds for us); regex ignores case, undocumented.

Should you consider it? When it is useful#

Consider it for a few simple guards on Bash commands and file edits on Mac or Linux, test every rule, and skip it where you need real security.

A decision flowchart for installing hookify: skip it if you do not want to block or flag commands or code, use a permission deny rule for one fixed command, skip it for now if you need prompt or stop rules or are on Windows without python3, otherwise install it for bash and file rules and test each rule once

Situation	Verdict
Block `git push --force` or warn on edits to `.env`	Install, test the rule
One fixed command to forbid	Use a permission deny rule
"Run the tests before you stop"	Skip for now; stop rules have open bugs
Windows without `python3`	Skip, or fix Python first

Install and check it#

You install it from Anthropic's official marketplace with one command. Check that python3 --version prints 3.9 or newer first.

Linux#

Run in terminalbash

claude plugin install hookify@claude-plugins-official

Windows (PowerShell)#

Reports say hookify needs a command called python3; many Windows installs only have python or py. Check first:

Run in terminalpowershell

python3 --version
claude plugin install hookify@claude-plugins-official

If the first command fails, expect the hooks to do nothing.

Mac (Terminal)#

Run in terminalbash

claude plugin install hookify@claude-plugins-official

Check it worked#

Inside Claude Code, in a scratch project, create a harmless test rule:

Codetext

/hookify Warn me when a command contains hookify-test

Then ask Claude to run echo hookify-test. The warning should appear on your screen. Then run claude plugin details hookify.

Things to keep in mind#

Security. The hooks run Python with your user rights on every tool call. Rule files run no code, but a cloned repository can carry some that block your tool calls.
Not a security boundary. Use permission deny rules, and the sandbox for real isolation.
Updates. No version number, so you cannot pin one.

Troubleshooting#

Problem	Solution
Rule never fires	The file must be `.claude/hookify.<name>.local.md` in the project, start Claude Code from the project root, and set `enabled: true`.
`Hookify error: Invalid \escape: line 1 column 467 (char 466)` (Windows)	Known backslash-path bug (issue 5730); no fix in the code we read.
`Hookify import error: No module named 'hookify'`	An old bug fixed in current code: update the plugin.
`PreToolUse:Bash hook error: Blocked by hook`	Normal for a block rule, but the rule name is not shown: check your `.claude/hookify.*.local.md` files.
Rule with emoji or accents is ignored on Windows	Keep rule files plain ASCII (issue 6092).

FAQ#

Is hookify made by Anthropic? Yes, with the "Anthropic verified" label.

How do I turn a rule off? Set enabled: false, delete the file, or run /hookify:configure.

Sources#

anthropics/claude-plugins-official, plugins/hookify: every file read at commit 41f619d (27 October 2026); last folder change b58bdbf (May 2026); marketplace entry in .claude-plugin/marketplace.json.
Hookify on the Claude marketplace: install count 60,376 and description (October 2026).
Hooks reference, Configure permissions and Install and manage plugins: official Claude Code docs (October 2026).
User reports: issues 20747, 99221, 93028, 78490, 79143 (2026) and 85, 292, 1788, 1859, 5368, 5660, 5662, 6144, 6368 (2025 to 2026); two Hacker News comments (February 2026).

Not run inside Claude Code by DIY Lab Hub: we read the files and docs and ran the hook scripts from a shell on sample rules, but did not install the plugin in a live session.

Checked against the official docs on 2026-10-27

Hookify for Claude Code: review, how it works and whether it is worth it (2026)

Before you start