Review an agent, skill, MCP server or workflow permission set for unnecessary access, privilege expansion and dangerous combinations before it is enabled.
Install
npx skillscat add molotroka123-cell/aimaxbossman/permission-auditor Install via the SkillsCat registry.
SKILL.md
Permission Auditor
Use before:
- enabling a new MCP server
- assigning new tools
- applying natural-language permission changes
- promoting a skill that expands permissions
- granting host filesystem or terminal access
- increasing browser action authority
Goal
Apply least privilege.
The question is not:
Could the agent use this permission?
It is:
Does this exact workflow require this permission?
Review categories
- filesystem read
- filesystem write
- external directories
- terminal sandbox
- host terminal
- administrator/sudo
- network
- browser navigation
- login
- upload/download
- external submit/send
- git push
- secrets
- MCP tools
- destructive actions
Dangerous combinations
Pay special attention to combinations such as:
- arbitrary filesystem read + external network
- secrets access + browser/send
- admin terminal + untrusted external content
- MCP filesystem + unrestricted shell
- git credentials + unreviewed push
- browser login + automatic external submit
Workflow
- Read task/skill requirements.
- List requested permissions.
- Mark each:
- REQUIRED
- OPTIONAL
- UNNECESSARY
- DANGEROUS
- Reduce permission scope:
- narrower path
- narrower domain
- specific command pattern
- specific MCP tool
- Choose:
- AUTO
- ASK
- DENY
- Compare with previous policy.
- Highlight every privilege increase.
Rules
Permission expansion must never be hidden inside a skill update.
Do not solve tool failure by broadening permissions unless the access is actually required.
Output
Subject:
Required permissions:
Removed permissions:
Privilege increases:
Dangerous combinations:
Recommended AUTO:
Recommended ASK:
Recommended DENY:
Human approval required: yes/no