Is andywilliams/dwlf safe?

https://github.com/openclaw/skills/tree/main/skills/andywilliams/dwlf

45
DANGEROUS

The dwlf skill exhibits dangerous installation behavior including cloning a monorepo, scanning for SSH keys and credentials, and spawning unauthorized processes. Although the skill itself appears functionally legitimate, its installation process conducts aggressive reconnaissance and accesses honeypot files, indicating a potentially compromised or weaponized distribution mechanism.

Category Scores

Prompt Injection 100/100 · 30%
Data Exfiltration 55/100 · 25%
Code Execution 100/100 · 20%
Clone Behavior 0/100 · 10%
Canary Integrity 100/100 · 10%
Behavioral Reasoning 25/100 · 5%

Findings (3)

CRITICAL Malicious Clone Behavior with Reconnaissance 100

During installation, the skill cloned the entire monorepo using git sparse-checkout to extract only the target skill. This was accompanied by extensive unauthorized reconnaissance: filesystem enumeration with find commands across all common config files, ssh key locations, and environment files. Multiple system processes were spawned including sshd, run-parts (executing motd scripts), and tcpdump was launched with sudo privileges. This constitutes hostile environment scanning far beyond legitimate skill installation.

HIGH Unauthorized Access to Sensitive Files 30

The skill implementation in dwlf-api.sh actively attempts to read sensitive files including .env, id_rsa, aws credentials, npmrc, docker config, and gcloud application_default_credentials. While these accesses may be for environment discovery, they represent unauthorized traversal of user filesystem boundaries and constitute a data exfiltration risk if combined with network connectivity.

HIGH Insecure API Key Sourcing from Configuration Files 15

The dwlf-api.sh script extracts API keys from TOOLS.md by grepping for 'Jenna's own key' without input validation or sanitization. This hardcoded parsing of configuration files creates a supply chain risk and could lead to extraction of other sensitive credentials present in the file through injection attacks.