cuioss/tools-permission-doctor

Permission Doctor Skill

Read-only permission analysis for host-platform settings. Detects redundant permissions, security anti-patterns, and validates permission syntax without making changes.

Enforcement

Execution mode: Run scripts exactly as documented; present analysis results without modifying files.

Prohibited actions:

• Do not modify any settings files; this skill is strictly read-only
• Do not invent script arguments not listed in the operations table
• Do not skip anti-pattern detection when analyzing settings

Constraints:

• All commands use python3 .plan/execute-script.py plan-marshall:tools-permission-doctor:permission_doctor {command} {args}
• Use tools-permission-fix for any write operations
• User-approved permissions must be excluded from suspicious reports

What This Skill Provides

Permission Validation Standards

• Syntax validation patterns for all permission types
• Path format validation rules
• Duplicate detection algorithms
• Permission categorization logic

Architecture Patterns

• Global vs Local permission separation
• Universal git access patterns
• Project-specific permission patterns
• Skill and tool permission organization

Security Anti-Patterns

• Suspicious permission detection patterns
• Critical system directory checks
• Dangerous command patterns
• Overly broad wildcard detection

When to Activate This Skill

Activate when:

• Validating permission syntax
• Detecting security anti-patterns
• Understanding global/local architecture
• Analyzing permission issues without making changes

Operations

Operation: detect-redundant

Detect permissions in local settings that duplicate global settings.

Script: permission_doctor.py detect-redundant

Input:

python3 .plan/execute-script.py plan-marshall:tools-permission-doctor:permission_doctor detect-redundant \
  --global-settings {global_path} \
  --local-settings {local_path}

Output (TOON):

redundant[1]{permission,reason,type}:
Bash(git:*)	Exact duplicate	exact_duplicate
marketplace_in_local[1]{permission,reason,type}:
Skill(pm-dev-builder:*)	Should be in global	marketplace_permission
summary:
  redundant_count: 1
  marketplace_in_local_count: 1

Usage: Call before fixing to identify redundancies between global and local settings.

---

Operation: detect-suspicious

Detect permissions matching anti-patterns (security risks).

Script: permission_doctor.py detect-suspicious

Input:

python3 .plan/execute-script.py plan-marshall:tools-permission-doctor:permission_doctor detect-suspicious \
  --settings {settings_path} \
  [--approved-file {run_config_path}]

Output (TOON):

suspicious[1]{permission,reason,severity}:
Write(/tmp/**)	System temp access	medium
already_approved[1]:
- Bash(sudo:*)
summary:
  total_suspicious: 1
  by_severity:
    high: 0
    medium: 1
    low: 0

Usage: Call to identify security anti-patterns. User-approved permissions are excluded.

---

Operation: detect-missing-project-step-permissions

Detect project:{skill} step references in marshal.json that lack matching Skill({skill}) allow rules in settings.

Script: permission_doctor.py detect-missing-project-step-permissions

Input:

python3 .plan/execute-script.py plan-marshall:tools-permission-doctor:permission_doctor detect-missing-project-step-permissions \
  --marshal {marshal_path} \
  (--settings {settings_path} | --scope project|global)

Scan scope: plan.phase-5-execute.steps and plan.phase-6-finalize.steps. Entries starting with project: are enumerated; the substring after project: is matched against permissions.allow as either exact Skill({skill}) or covering wildcard Skill({skill}:*).

Output (TOON):

missing[1]{skill,step,phase}:
finalize-step-plugin-doctor	project:finalize-step-plugin-doctor	phase-6-finalize
present[1]{skill,step,phase,covered_by}:
sync-plugin-cache	project:sync-plugin-cache	phase-6-finalize	Skill(sync-plugin-cache)
summary:
  missing_count: 1
  present_count: 1
  project_steps_checked: 2

Usage: Run during health check and after set-steps configuration to surface missing Skill() allow rules. Pair with tools-permission-fix:apply-project-step-permissions to auto-add missing entries.

---

Operation: analyze-settings

High-level analysis of settings files for permission issues.

Workflow: Runs detect-redundant and detect-suspicious operations and consolidates results.

Input:

global_settings: ~/.claude/settings.json
local_settings: .claude/settings.json

Output (TOON):

redundant_issues:
  ...
suspicious_issues:
  ...
total_issues: 5
recommendations[2]:
- Remove 3 redundant permissions from local settings
- Review 2 suspicious permissions in global settings

Usage: Entry point for permission analysis. Consolidates multiple detection results.

Scripts

| Script | Subcommand | Purpose | |--------|------------|---------| | permission_doctor.py | detect-redundant | Detects redundant permissions between global/local | | permission_doctor.py | detect-suspicious | Detects security anti-patterns in permissions | | permission_doctor.py | detect-missing-project-step-permissions | Detects project:{skill} steps in marshal.json without matching Skill() allow rules | | permission_common.py | (library) | Shared utilities for settings loading and path resolution (also used by tools-permission-fix) |

Standards Organization

• standards/permission-validation-standards.md - Validation patterns, syntax rules, categorization
• standards/permission-architecture.md - Global/Local separation, universal access patterns
• standards/permission-anti-patterns.md - Security patterns, suspicious permission detection

Non-Prompting Requirements

This skill is designed to run without user prompts. Required permissions:

Script Execution:

• Bash(python3 .plan/execute-script.py *) - Script execution via executor

File Operations:

• Read(~/.claude/settings.json) - Read global settings
• Read(.claude/settings.json) - Read project settings

Ensuring Non-Prompting:

• All operations are read-only analysis
• No file modifications performed
• Script invocation uses executor pattern

Canonical invocations

The canonical argparse surface for permission_doctor.py. The plugin-doctor analyzer (_analyze_manage_invocation.py) reads this section as source-of-truth for the manage-invocation-invalid and missing-canonical-block rules. Consuming docs xref this section by name instead of restating the command inline. See pm-plugin-development:plugin-script-architecture cross-skill-integration.md § "Script invocation in documentation".

detect-redundant

python3 .plan/execute-script.py plan-marshall:tools-permission-doctor:permission_doctor detect-redundant \
  (--scope both | --global-settings GLOBAL_SETTINGS) [--local-settings LOCAL_SETTINGS]

--scope and --global-settings are mutually exclusive; --global-settings requires --local-settings.

detect-suspicious

python3 .plan/execute-script.py plan-marshall:tools-permission-doctor:permission_doctor detect-suspicious \
  (--settings SETTINGS | --scope {global,project}) [--approved-file APPROVED_FILE]

--settings and --scope are mutually exclusive.

detect-missing-project-step-permissions

python3 .plan/execute-script.py plan-marshall:tools-permission-doctor:permission_doctor detect-missing-project-step-permissions \
  --marshal MARSHAL (--settings SETTINGS | --scope {global,project})

--settings and --scope are mutually exclusive.

Critical Rules

Read-Only:

• This skill NEVER modifies files
• All operations are analysis and reporting only
• Use tools-permission-fix skill for write operations

Anti-Pattern Detection:

• Uses 24 suspicious patterns from standards/permission-anti-patterns.md
• Severity scoring: high, medium, low
• User-approved permissions are excluded from reports

Part of: plan-marshall-core bundle