Skill Permissions Skill

Skill Permissions

Automatically analyze permissions required by all skills in Claude Code and generate one-time authorization commands.

Dangerous Operations Warning

This skill can modify other skill files. The following commands may cause widespread file changes:

| Command | Risk Level | Description | |---------|------------|-------------| | /skill-permissions inject | High | Batch modify all skill SKILL.md files | | /skill-permissions allow-all | Medium | Batch modify settings.json | | /skill-permissions allow <name> | Low | Only modify settings.json | | /skill-permissions <name> | Safe | Read-only analysis, no file modifications |

Recommendations:

Prefer explicit commands (e.g., /skill-permissions allow port-allocator)
Avoid inject or allow-all unless you understand the impact
Run /skill-permissions first to see analysis results before batch operations

Usage

| Command | Description | |---------|-------------| | /skill-permissions | Analyze all skills and show permission summary | | /skill-permissions <skill-name> | Analyze specific skill's permission requirements | | /skill-permissions allow <skill-name> | Execute one-time authorization for specific skill | | /skill-permissions allow-all | Execute one-time authorization for all skills | | /skill-permissions inject | Inject allow command into all skills | | /skill-permissions allow | Execute one-time authorization for this skill |

How It Works

1. Permission Analysis

Scan skill's SKILL.md file, identifying patterns like:

Bash Command Patterns:

find * → Bash(find *)
ls * → Bash(ls *)
cat ~/.claude/* → Bash(cat ~/.claude/*)
lsof -i:3* → Bash(lsof -i:3*)
git * → Bash(git *)
npm * → Bash(npm *)

Recognition Rules:

Bash commands in code blocks (bash ... )
Inline commands (command)
Explicitly mentioned system commands

2. Permission Rule Generation

Convert identified commands to permissions.allow rules:

{
  "permissions": {
    "allow": [
      "Bash(find * -name package.json *)",
      "Bash(ls -d *)",
      "Bash(cat ~/.claude/*)",
      "Bash(lsof -i:3*)"
    ]
  }
}

Execution Steps

Command: `/skill-permissions`

Analyze all skills and show permission summary:

Scan all skills

find ~/.claude/skills -name "SKILL.md" -type f 2>/dev/null

Analyze permissions for each skill
- Read SKILL.md content
- Extract bash code blocks
- Identify command patterns
- Generate permission rules
Output analysis results

Command: `/skill-permissions <skill-name>`

Analyze specific skill's permission requirements:

Locate skill

SKILL_PATH=$(find ~/.claude/skills -type d -name "<skill-name>" 2>/dev/null | head -1)

Read and analyze SKILL.md
Output detailed permission list

Command: `/skill-permissions allow <skill-name>`

Execute one-time authorization for specific skill:

Analyze skill permission requirements
Read existing config
Merge permission rules (deduplicate, preserve existing rules)
Write config file
Output authorization result

Blocked Commands

The following command patterns are automatically blocked and won't be added to the allowlist:

Dangerous File Operations

| Command | Reason | |---------|--------| | rm * | File deletion, may cause data loss | | rm -rf * | Recursive force delete, extremely dangerous | | sudo * | Super user privileges | | chmod 777 * | Opens all permissions |

Dangerous Process Operations

| Command | Reason | |---------|--------| | kill -9 * | Force kill process | | pkill * | Kill processes by name | | curl * \| bash | Remote code execution | | eval * | Dynamic code execution |

Dangerous Git Operations

| Command | Reason | |---------|--------| | git push --force * | Force push, may overwrite remote | | git reset --hard * | Hard reset, loses uncommitted changes |

Output Format

Analysis Result (Single Skill)

Skill: port-allocator
Location: ~/.claude/skills/port-allocator

Detected commands:
  - find ~/Codes -maxdepth 3 -name "package.json"
  - ls -d */
  - cat ~/.claude/port-registry.json
  - lsof -i:3000

Generated permission rules:
  - Bash(find * -maxdepth * -name package.json *)
  - Bash(ls -d *)
  - Bash(cat ~/.claude/*)
  - Bash(lsof -i:3*)

Run `/skill-permissions allow port-allocator` to authorize

Authorization Success

Permission authorization successful

Skill: port-allocator

Added permission rules:
  - Bash(find * -maxdepth * -name package.json *)
  - Bash(ls -d *)
  - Bash(cat ~/.claude/*)
  - Bash(lsof -i:3*)

Config file: ~/.claude/settings.json

New permissions take effect in next session, or run /clear to apply immediately

Configuration Files

Permission config: ~/.claude/settings.json
Skill directory: ~/.claude/skills/

First Use

If you encounter permission prompts, first run:

/skill-permissions allow

Command: `/skill-permissions allow`

Execute one-time authorization for this skill:

Read ~/.claude/settings.json
Merge the following permissions to permissions.allow:

{
  "permissions": {
    "allow": [
      "Bash(find ~/.claude/skills *)",
      "Bash(cat ~/.claude/*)"
    ]
  }
}

Write config file (preserve existing permissions)
Output authorization result

Notes

Conservative analysis - Only identify explicitly appearing commands, avoid over-authorization
Deduplicated merge - New permissions merge with existing, no duplicates
No deletion - Only adds permissions, won't delete user's existing permission config
Wildcards - Use * to match varying parameter parts
Session effect - Permission updates require new session or /clear to take effect
Explicit preferred - Recommend explicit commands over batch operations to reduce risks

Agent Skills: Skill Permissions

Install this agent skill to your local

Skill Files