auto-permissions-review-install
Install an AI-powered permission review hook that uses Claude Haiku to evaluate tool call safety before execution. Per-session, auto-approves read-only operations, sends ambiguous Bash commands to Haiku, and only reviews edits in accept-edits mode. Use when setting up auto-approve with AI review, reducing permission fatigue, or wanting Claude to review its own tool calls.
Best use case
auto-permissions-review-install is best used when you need a repeatable AI agent workflow instead of a one-off prompt.
Install an AI-powered permission review hook that uses Claude Haiku to evaluate tool call safety before execution. Per-session, auto-approves read-only operations, sends ambiguous Bash commands to Haiku, and only reviews edits in accept-edits mode. Use when setting up auto-approve with AI review, reducing permission fatigue, or wanting Claude to review its own tool calls.
Teams using auto-permissions-review-install should expect a more consistent output, faster repeated execution, less prompt rewriting.
When to use this skill
- You want a reusable workflow that can be run more than once with consistent structure.
When not to use this skill
- You only need a quick one-off answer and do not need a reusable workflow.
- You cannot install or maintain the underlying files, dependencies, or repository context.
Installation
Claude Code / Cursor / Codex
Manual Installation
- Download SKILL.md from GitHub
- Place it in
.claude/skills/auto-permissions-review-install/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How auto-permissions-review-install Compares
| Feature / Agent | auto-permissions-review-install | Standard Approach |
|---|---|---|
| Platform Support | Not specified | Limited / Varies |
| Context Awareness | High | Baseline |
| Installation Complexity | Unknown | N/A |
Frequently Asked Questions
What does this skill do?
Install an AI-powered permission review hook that uses Claude Haiku to evaluate tool call safety before execution. Per-session, auto-approves read-only operations, sends ambiguous Bash commands to Haiku, and only reviews edits in accept-edits mode. Use when setting up auto-approve with AI review, reducing permission fatigue, or wanting Claude to review its own tool calls.
Where can I find the source code?
You can find the source code on GitHub using the link provided at the top of the page.
SKILL.md Source
# Auto Permissions Review
Installs a `PreToolUse` hook that reduces permission prompt fatigue by auto-approving safe operations and sending ambiguous commands to Claude Haiku for review.
## Behavior
| Tool | Default mode | Accept-edits mode |
|------|-------------|-------------------|
| `Read`, `Glob`, `Grep`, `LS`, etc. | instant allow | instant allow |
| Simple Bash (`ls`, `cat`, `find`, `git status`) | instant allow | instant allow |
| Complex Bash (pipes, substitution) | Haiku reviews | Haiku reviews |
| `Edit`, `Write` | normal prompt (you decide) | Haiku reviews |
Key design choices:
- **Per-session** -- each terminal session independently enables/disables. Multiple sessions can run simultaneously without conflicts.
- **Read-only operations never prompt** -- `Read`, `Glob`, `Grep`, and common Bash reads are auto-approved instantly with no API call.
- **Ambiguous Bash always goes to Haiku** -- piped commands, subshells, and anything not in the read-only pattern list gets reviewed regardless of mode.
- **Edits respect your permission mode** -- in default mode you still approve edits manually. In accept-edits mode (Shift+Tab), Haiku reviews them.
## Related Commands
| Command | What It Does |
|---------|--------------|
| `/auto-permissions-review-install` | Install the hook (run once) |
| `/auto-permissions-review-enable` | Turn on for this session |
| `/auto-permissions-review-disable` | Turn off for this session |
## How It Works
1. User runs `/auto-permissions-review-enable` -- hook creates `~/.claude/ai-review-sessions/<session_id>`
2. Tool call triggers `PreToolUse` hook
3. Hook checks if a file for this session's ID exists
4. Read-only tools and simple Bash reads are auto-approved instantly
5. Complex Bash commands go to `claude -p --model claude-haiku-4-5-20251001 --permission-mode plan`
6. Write tools check permission mode -- Haiku in accept-edits, normal prompt in default
7. Haiku responds `ALLOW` or `BLOCK: <reason>`
8. Hook returns the decision to Claude Code
## Installation
Run the install script to set everything up:
```bash
bash "${CLAUDE_PLUGIN_ROOT}/skills/auto-permissions-review-install/scripts/install.sh"
```
This will:
1. Copy `ai-review.sh` to `~/.claude/hooks/`
2. Register the `PreToolUse` hook in `~/.claude/settings.json`
## Workflow
### Step 1: Check If Already Installed
```bash
[ -f ~/.claude/hooks/ai-review.sh ] && echo "Already installed" || echo "Not installed"
```
If already installed, tell the user and ask if they want to reinstall.
### Step 2: Run the Installer
```bash
bash "${CLAUDE_PLUGIN_ROOT}/skills/auto-permissions-review-install/scripts/install.sh"
```
### Step 3: Confirm and Explain
After installation, tell the user:
```
Auto Permissions Review hook installed.
Usage:
/auto-permissions-review-enable -- turn on (this session only)
/auto-permissions-review-disable -- turn off (this session only)
Read-only operations are auto-approved instantly.
Ambiguous Bash commands are reviewed by Haiku.
Edits only go through Haiku when accept-edits mode is on (Shift+Tab).
Each session enables independently.
```
## Customization
The hook script is at `~/.claude/hooks/ai-review.sh`. Users can:
- Change the model (swap `claude-haiku-4-5-20251001` for any model)
- Add patterns to the read-only Bash allowlist
- Modify the Haiku evaluation prompt/rules
- Set `DEBUG=0` to disable logging
## Debug Log
When `DEBUG=1` (default), all hook activity is logged to `~/.claude/ai-review.log`:
```bash
cat ~/.claude/ai-review.log
```
## Requirements
- Claude Code CLI (`claude` command available)
- `jq` installed (`brew install jq`)Related Skills
review-analyst-agent
Use this skill to analyze product reviews, find common issues, and prioritize improvements. Triggers: "analyze reviews", "review analysis", "customer feedback", "what are people saying", "product reviews", "review sentiment", "find complaints", "customer complaints", "improvement recommendations", "voice of customer", "VOC analysis", "feedback analysis" Outputs: Prioritized issues, sentiment analysis, improvement recommendations.
auto-permissions-review-enable
Enable the auto permissions review hook for this session. Auto-approves read-only operations, sends ambiguous Bash to Haiku, and reviews edits only in accept-edits mode. Other terminal sessions are not affected.
auto-permissions-review-disable
Disable the auto permissions review hook for this session. Tool calls return to normal permission prompts. Other terminal sessions are not affected.
xlsx
Comprehensive spreadsheet creation, editing, and analysis with support for formulas, formatting, data analysis, and visualization. When Claude needs to work with spreadsheets (.xlsx, .xlsm, .csv, .tsv, etc) for: (1) Creating new spreadsheets with formulas and formatting, (2) Reading or analyzing data, (3) Modify existing spreadsheets while preserving formulas, (4) Data analysis and visualization in spreadsheets, or (5) Recalculating formulas
walkthrough-script-agent
Use this skill to generate walkthrough video scripts for app features. Triggers: "walkthrough script", "feature walkthrough", "app walkthrough script", "demo script", "video script for app", "feature tour script", "product walkthrough", "screen-by-screen script", "app feature scripts", "walkthrough narration", "tutorial script", "onboarding script", "feature highlight script" Outputs: Timed walkthrough scripts with narration, screen directions, and transitions for each app feature.
voice-generation
Use this skill for AI text-to-speech generation. Triggers include: "generate voice", "create audio", "text to speech", "TTS", "read this aloud", "generate narration", "create voiceover", "synthesize speech", "podcast audio", "dialogue audio", "multi-speaker", "audiobook" Supports Google Gemini TTS, ElevenLabs, and OpenAI TTS.
video-producer-agent
Use this skill to create complete videos with voiceover and music. Triggers: "create video", "product video", "explainer video", "promo video", "demo video", "training video", "ad video", "commercial", "marketing video", "video with voiceover", "video with music", "brand video", "testimonial video" Orchestrates: script, voiceover, background music, video clips/images, and final assembly.
video-generation
Use this skill for AI video generation. Triggers include: "generate video", "create video", "make video", "animate", "text to video", "video from image", "video of", "animate image", "bring to life", "make it move", "add motion", "video with audio", "video with dialogue" Supports text-to-video, image-to-video, video with dialogue/audio using Google Veo 3.1 (default) or OpenAI Sora.
style-guide
Analyze a codebase to extract its conventions, patterns, and style. Spawns specialized analyzer agents that each focus on one aspect (structure, naming, patterns, testing, frontend). Generates a comprehensive style guide that other skills can reference. Use when starting work on an unfamiliar codebase, or to create explicit documentation of implicit conventions.
social-producer-agent
Use this skill to create multiple social media assets as a coordinated pack. Triggers: "social media content", "content pack", "social assets", "campaign assets", "instagram content", "tiktok content", "launch kit", "marketing kit", "content series", "social media kit", "multiple posts", "content calendar", "batch content" Orchestrates: multiple images, short videos, and audio for social platforms.
slide-generation
Use this skill to create presentation slides from structured content. Triggers: "create slides", "generate presentation", "make powerpoint", "create pptx", "build slides", "presentation from content", "slide deck", "slides for" Outputs: PPTX files, Markdown slides, or HTML presentations. Used by: pitch-deck-agent, market-researcher-agent, and other agents needing slides.
sidequest
Spawn a new Claude Code session in a separate terminal to work on a different task, optionally with context from the current session. Use when you need to work on something else without losing your current progress.