feat(claude): enhance permissions with context-aware safety patterns

[technicalpickles]

Summary

Comprehensive improvements to Claude Code permissions, balancing security with usability through a three-tier system: allow safe patterns, ask for moderate risk, deny catastrophic operations.

New Features

New Ecosystem Files:

• permissions.beans.json - Beans issue tracker commands
• permissions.colima.json - Colima container management

New Skill:

• permissions-manager - Local skill for analyzing and managing permissions
  • /permissions-manager analyze - Generate recommendations
  • /permissions-manager apply - Interactively apply changes
  • /permissions-manager review - Quick status overview

Key Improvements

Safe rm -rf Patterns:

• Allow safe build artifacts: node_modules, dist, build, target, etc.
• Use exact matches without wildcards to prevent multi-path deletions
• Added both relative (dist) and explicit (./dist) variants

Granular sudo Permissions:

• Allow: systemctl, journalctl, docker (service management)
• Ask: chmod, chown, shutdown, package managers (moderate risk)
• Deny: rm -rf, dd, mkfs, fdisk (destructive operations)

Git Force-Push Protection:

• Explicitly deny force-push to main/master branches
• Allow safer alternative: --force-with-lease
• Changed force-push to feature branches from deny → ask
• Added common safe reset patterns

Shell Utilities:

• Added 30+ common utilities: bash, awk, rsync, ssh, ps, pkill, sqlite3, yamllint, etc.

Other Changes:

• Added documentation domains for WebFetch (mise, karafka, lima-vm)
• Changed MCP destructive operations from deny → ask
• Added work-specific bktide commands

Security Enhancements

Wildcard Safety:

• Removed dangerous :* wildcards from rm -rf commands
• Prevents patterns like rm -rf node_modules /etc from matching
• Kept wildcards only for truly safe operations

Explicit Denies:

• Absolute path deletion: /, /*, ~, ~/*, $HOME
• Parent directory deletion: ., .., ../*
• Pipe-to-shell: curl | bash, wget | sh
• Protected branches: force-push to main/master

Documentation

Analysis & Guides:

• claude-permissions-analysis.md - Full analysis and recommendations
• claude-permissions-safe-patterns.md - Context-specific safety patterns
• claude-permissions-wildcard-safety.md - Wildcard security analysis
• claude-permissions-changes-summary.md - Complete change summary

Privacy Features:

• Permissions-manager skill auto-redacts project names
• Timestamped documentation safe for commits
• Sanitizes paths and domains

Stats

Metric	Before	After	Change
Allow	346	423	+77 (+22%)
Ask	2	25	+23 (+1150%)
Deny	9	25	+16 (+178%)

Cleanup

Removed 37 duplicate permission entries from 16 project files.

Test Plan

• Tested rm -rf patterns prevent multi-path deletions
• Verified safe patterns auto-allow (no prompts)
• Confirmed dangerous patterns prompt or deny
• Regenerated settings successfully
• Cleaned up project duplicates

Commits

• feat(claude): enhance permissions with context-aware safety patterns
• feat(claude): add permissions-manager skill

🤖 Generated with Claude Code