
aaw-security
by jieyao-MilestoneHub
Multi-agent AI workspace with role isolation and auditable contract governance (with Claude Code).
SKILL.md
name: aaw-security description: Enforces security rules for AAW sessions. Activated automatically when handling sensitive operations, external connections, package installations, or file access to protected paths. allowed-tools:
- mcp__aawctl__exec
- mcp__aawctl__read_policy
- Read
- Glob
- Grep
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 commandssudoanythingchmod 777or overly permissive permissions- Commands that modify system files
- Commands that disable security features
3. External Connections
NEVER do:
curlorwgetto 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:
-
Check approval status
- Is the package in the approved list?
- Is it from a trusted registry?
-
If not approved:
- STOP immediately
- Report the package name and version
- Wait for explicit approval
-
Log the installation
- Use
mcp__aawctl__execfor npm/yarn/pnpm - Provide purpose: "Installing package X for Y feature"
- Use
File Access
When accessing files:
-
Check against deny patterns:
.env*- DENIEDsecrets/**- DENIED**/credentials*- DENIED**/*.pem,**/*.key- DENIED
-
If denied:
- Do not attempt to read
- Report the access attempt
- Suggest alternatives
External API Calls
When making external calls:
- Use approved endpoints only
- All calls must be logged through aawctl
- No direct curl/wget
Incident Response
If you detect a potential security violation:
- STOP - Do not proceed with the action
- LOG - Record the incident details
- REPORT - Inform the user immediately
- 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 |
Score
Total Score
Based on repository quality metrics
SKILL.mdファイルが含まれている
ライセンスが設定されている
100文字以上の説明がある
GitHub Stars 100以上
3ヶ月以内に更新
10回以上フォークされている
オープンIssueが50未満
プログラミング言語が設定されている
1つ以上のタグが設定されている
Reviews
Reviews coming soon
