aaw-security

jieyao-MilestoneHub's avatarfrom jieyao-MilestoneHub

Enforces security rules for AAW sessions. Activated automatically when handling sensitive operations, external connections, package installations, or file access to protected paths.

⭐1stars🔀0forks📁View on GitHub🕐Updated Jan 9, 2026
[Guardrails and Safety]

When & Why to Use This Skill

The aaw-security skill serves as a comprehensive security framework for Claude AAW sessions, implementing strict guardrails to prevent unauthorized actions. It automatically monitors and restricts access to sensitive files, blocks dangerous system commands, and manages external connections to ensure a secure and compliant agentic environment. By enforcing predefined security policies, it mitigates risks of data leakage and system compromise during automated tasks.

Use Cases

  • Preventing accidental exposure of sensitive credentials, .env files, and private keys during code analysis or file processing.
  • Blocking high-risk system operations such as 'sudo' commands, destructive file deletions, or unauthorized permission changes.
  • Securing the software supply chain by validating package installations against approved registries and logging all deployment activities.
  • Enforcing network safety by restricting direct external API calls and ensuring all outbound traffic goes through authorized controllers.
  • Providing real-time incident response and audit logging for security-sensitive operations to maintain organizational compliance.
nameaaw-security
descriptionEnforces security rules for AAW sessions. Activated automatically when handling sensitive operations, external connections, package installations, or file access to protected paths.

AAW Security Rules

This skill enforces security policies for all AAW operations.

Absolute Prohibitions

1. Secrets and Credentials

NEVER do these:

  • Read .env, .env.*, or any environment files
  • Read files in secrets/ directories
  • Read files named credentials*, *.pem, *.key
  • Log, output, or display secret values
  • Commit secrets to version control
  • Store secrets in code comments

If you need secrets:

  • Ask the user to provide them securely
  • Use environment variables at runtime
  • Reference secret management systems

2. Dangerous Commands

NEVER execute:

  • rm -rf / or similar destructive commands
  • sudo anything
  • chmod 777 or overly permissive permissions
  • Commands that modify system files
  • Commands that disable security features

3. External Connections

NEVER do:

  • curl or wget to unknown domains
  • Direct API calls to external services (use aawctl)
  • Download and execute scripts from the internet
  • Connect to databases without going through aawctl

Controlled Operations

Package Installation

When installing packages:

  1. Check approval status

    • Is the package in the approved list?
    • Is it from a trusted registry?

  2. If not approved:

    • STOP immediately
    • Report the package name and version
    • Wait for explicit approval

  3. Log the installation

    • Use mcp__aawctl__exec for npm/yarn/pnpm
    • Provide purpose: "Installing package X for Y feature"

File Access

When accessing files:

  1. Check against deny patterns:

    • .env* - DENIED
    • secrets/** - DENIED
    • **/credentials* - DENIED
    • **/*.pem, **/*.key - DENIED

  2. If denied:

    • Do not attempt to read
    • Report the access attempt
    • Suggest alternatives

External API Calls

When making external calls:

  1. Use approved endpoints only
  2. All calls must be logged through aawctl
  3. No direct curl/wget

Incident Response

If you detect a potential security violation:

  1. STOP - Do not proceed with the action
  2. LOG - Record the incident details
  3. REPORT - Inform the user immediately
  4. WAIT - Do not continue without explicit approval

Security Logging

All security-relevant events are logged:

  • Policy check results
  • Denied access attempts
  • Package installations
  • External connections

Logs are included in the evidence bundle.

Policy Enforcement Modes

Audit-Only Mode (Default for PoC)

  • Violations are LOGGED but not BLOCKED
  • Tagged as policy_check: WOULD_BLOCK
  • Use for initial rollout and tuning

Enforce Mode (Production)

  • Violations are BLOCKED
  • Operation fails with error message
  • Requires explicit override (if allowed)

Quick Reference

Action Rule
Read .env DENIED
Run sudo DENIED
curl external DENIED
npm install (approved) ALLOWED
npm install (unknown) ASK
Edit source files ALLOWED
Edit node_modules DENIED