AOI OpenClaw Security Toolkit (Core)

Why: Prevent “one bad commit” incidents (accidental file leakage + secret exposure) with a fast, local-only, fail-closed check.

When: Before committing/pushing, before publishing a skill, and when reviewing scripts/skills for unexpected egress behavior.

How: Run a single command to get PASS/WARN/BLOCK and an optional redaction-safe report.

Scope: Detection + reporting only (no auto-fix, no uploads, no auto-posting).

Quickstart: openclaw-sec check --preset repo --diff staged

This is a public-safe toolkit skill.

• Does: detect + report risks (PASS/WARN/BLOCK)
• Does NOT: auto-fix, auto-upload, auto-post, or exfiltrate data

CLI

Binary: openclaw-sec

Common:

openclaw-sec check --lang en
openclaw-sec check --lang ko
openclaw-sec scan-secrets
openclaw-sec scan-egress
openclaw-sec scan-prompt --file inbound.txt

Exit codes:

• 0 PASS
• 1 WARN
• 2 BLOCK

Default scan scope

If --paths is omitted, it scans existing paths among:

• .
• skills/
• scripts/
• context/

Rules

Rule files live in rules/:

• secret_patterns.txt
• egress_patterns.txt
• prompt_injection_patterns.txt

Edit these to tune sensitivity.