Is gsc safe?

https://clawhub.ai/jdrhyne/gsc

68
CAUTION

This skill is an empty shell — no SKILL.md content, no source code, no package.json — containing only a lock.json referencing a phantom 'academic-research-hub' dependency. Despite having no visible functionality, its installation triggered reads of .env, .aws/credentials, and agent auth-profiles, along with creation of 25+ transpiled runtime modules. While no outbound network exfiltration was detected, the pattern of accessing sensitive credential files without any declared purpose is highly suspicious and consistent with a reconnaissance probe.

Category Scores

Prompt Injection 70/100 · 30%
Data Exfiltration 45/100 · 25%
Code Execution 55/100 · 20%
Clone Behavior 65/100 · 10%
Canary Integrity 100/100 · 10%
Behavioral Reasoning 40/100 · 5%

Findings (7)

HIGH Sensitive credential files accessed during install -35

During skill installation, the process opened and read /home/oc-exec/.env and /home/oc-exec/.aws/credentials. These files contain API keys, secrets, and AWS access credentials. While no outbound network exfiltration was detected, reading these files is unnecessary for skill installation and represents a significant privacy/security concern.

HIGH Agent auth profiles accessed -20

The installation process read /home/oc-exec/.openclaw/agents/main/agent/auth-profiles.json, which contains agent authentication and authorization data. This access is not justified by any visible skill functionality.

MEDIUM Empty SKILL.md with phantom dependency -30

The skill has no SKILL.md content and no source code, but its lock.json references an 'academic-research-hub' dependency that is not present in the repository. This phantom dependency could be a mechanism for lazy-loading prompt injection content or code at runtime through the platform's dependency resolution.

MEDIUM Disproportionate runtime module loading -30

Installation triggered creation and execution of 25+ transpiled JavaScript modules in /tmp/jiti/ including agent runtime infrastructure (exec-safety, agent-scope, plugin-sdk, memory-core). This volume of code execution is wildly disproportionate for a skill that contains no visible source code or package.json.

MEDIUM Extensive out-of-directory file access during install -35

The installation process accessed numerous files outside the skill directory including shell profiles (.profile, .bashrc), platform config (openclaw.json read 7 times), system files (/etc/hosts, /etc/nsswitch.conf), and credential files. This breadth of filesystem access is abnormal for installing a skill with no code.

LOW Possible reconnaissance probe -60

The combination of an empty skill with no functionality that nonetheless reads sensitive credential files during installation suggests this may be a reconnaissance probe — testing the execution environment's file access controls, identifying credential locations, and mapping the platform's directory structure for future exploitation.

INFO Canary files intact 0

All honeypot/canary files remained unmodified, indicating no direct tampering with planted credentials. However, real credential files (.env, .aws/credentials) were accessed.