agentic-ai-engineering/products/security-foundation/skills/damage-control/SKILL.md

name	description	user-invocable
damage-control	Three-level access control for agent filesystem operations. Zero-access, read-only, and no-delete path rules.	true

Damage Control Rules

Apply three-level access control to protect sensitive files from agent operations.

Three Access Levels

Zero Access (can't read or write)

zeroAccessPaths:
  - .env
  - ~/.ssh/
  - *.pem
  - credentials.json

Read-Only (can read, can't modify)

readOnlyPaths:
  - package-lock.json
  - yarn.lock
  - node_modules/

No-Delete (can modify, can't delete)

noDeletePaths:
  - .git/
  - Dockerfile
  - README.md
  - .github/workflows/

Dangerous Patterns

bashToolPatterns:
  - pattern: "^rm -rf"
    block: true
  - pattern: "git reset --hard"
    block: true
  - pattern: "DROP DATABASE"
    block: true
  - pattern: "sudo"
    ask: true

Usage

# Save as .pi/damage-control-rules.yaml
pi -e extensions/damage-control.ts

$ARGUMENTS