Content Security Policy Skill Guide

What this skill does

Most skills focus on "what the agent can do." In real workflows, you also need:

• What it must not do
• What requires confirmation
• Who can authorize which actions

The content security policy skill helps enforce those boundaries.

Capabilities typically include:

• Sensitive information masking
• Dangerous command filtering
• Permission checks for operations
• Admin vs regular user rules

When to install

Install when:

• OpenClaw is in real group chats
• It reads/writes documents, files, and messages
• You want to prevent dangerous or sensitive actions
• You need separate rules for admins, authorized users, and regular users

Skip if you are still in day-one testing or Feishu is not yet stable.

Four risk categories to configure

1. Sensitive information

Examples:

• IP addresses
• API keys
• Access tokens
• Local file paths

Default behavior: mask or redact in outputs.

2. Forbidden high-risk commands

Examples:

• ssh, scp, rm -rf /
• Direct reads of .env or SSH private keys

These should be rejected at the policy layer, not left for ad hoc judgment.

3. Requires-confirmation actions

Examples:

• Modify config
• Uninstall skills
• Broadcast messages
• Output key-related information

Configure by user role: admins allowed, others blocked or require explicit approval.

4. Logged medium-risk actions

Examples:

• File reads
• Upload/download
• List operations

Do not always block, but log and consider rate limits.

Rule configuration patterns

Admin vs regular users

admin:
  - config:edit
  - skill:uninstall
  - message:broadcast
regular:
  - file:read
  - message:send
  - search

Define which roles can perform which action types.

High-risk interception

blockedCommands:
  - "rm -rf"
  - "ssh"
  - "scp"
maskPatterns:
  - "api_key"
  - "token"
  - "password"

Block commands and mask sensitive strings in outputs.

Output filtering

• Redact IPs, keys, and tokens before sending replies
• Flag suspicious content for human review
• Log actions that cross policy boundaries

Configuration tips

• Start with strict rules and relax only when justified
• Clearly document admin vs regular user boundaries
• Avoid rules that are too broad (over-blocking) or too narrow (missed cases)
• Run openclaw security audit --deep periodically

External links

• OpenClaw docs
• OpenClaw GitHub
• Security checklist
• Anti-patterns