skill-differ

name	skill-differ
description	Compare two versions of an OpenClaw skill to detect security-relevant changes. Use before updating any skill from ClawHub. Highlights new capabilities, changed patterns, and recommends whether an update is safe.
user-invocable	true
metadata	{"openclaw":{"emoji":"🔄","os":["darwin","linux"],"requires":{"bins":"[Truncated]"}}}

Skill Differ

Compare two versions of an OpenClaw skill to find security-relevant changes before updating.

Why This Exists

A skill that was clean at v1.0 could add credential stealing in v1.1. The skill scanner catches known bad patterns in a single version. The differ catches new capabilities between versions — things a skill couldn't do before but can do now.

Commands

Diff two skill directories

python3 {baseDir}/scripts/differ.py diff --old ~/.openclaw/skills/some-skill/ --new /tmp/some-skill-v2/

Diff with JSON output

python3 {baseDir}/scripts/differ.py diff --old ./v1/ --new ./v2/ --json

Quick summary only (no file details)

python3 {baseDir}/scripts/differ.py diff --old ./v1/ --new ./v2/ --summary

What It Detects

New Capabilities Added

• Network access (skill didn't make HTTP requests before, now it does)
• Credential access (didn't read env vars or API keys before, now it does)
• File system access (wasn't touching home directory, now it is)
• Code execution patterns (eval/exec that didn't exist before)
• Data exfiltration (new outbound POST requests)
• Obfuscation (new encoded/obfuscated content)

File Changes

• New files added (especially in scripts/)
• Deleted files (could remove safety checks)
• Modified files with security-relevant diffs

Recommendations

• SAFE — No new security-relevant capabilities. Update freely.
• REVIEW — New capabilities detected. Read the changes before updating.
• BLOCK — Critical new capabilities (code execution, credential access). Manual audit required.

Tips

• Always diff before updating any third-party skill
• Pair with skill-scanner: scan before first install, diff before every update
• Pay attention to new files — attackers add payloads in new scripts
• If a "bug fix" update adds network access, that's suspicious