skill-web-importer
Pass
Audited by Gen Agent Trust Hub on Apr 5, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill's reference documentation (ref/web-threat-patterns.md) contains various prompt injection strings such as 'ignore previous instructions' and 'disregard your guidelines'. These are used as regex patterns for the skill's detection engine and are not instructions for the agent to follow.
- [COMMAND_EXECUTION]: The skill documents dangerous command patterns like 'curl | bash' and 'eval()' to provide context for its security scan. These examples are part of a threat taxonomy and do not represent executable code within the skill itself.
- [DATA_EXFILTRATION]: Reference files mention sensitive paths like '
/.aws/credentials' and '/.ssh/id_rsa'. These are used to identify potential data exposure in external skills being audited and do not indicate that this skill accesses the user's private credentials. - [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill fetches untrusted content from external URLs via 'WebFetch'. This risk is addressed through robust instructions that command the agent to treat all fetched text as untrusted data and never to follow instructions found within scanned files.
- [REMOTE_CODE_EXECUTION]: Example exfiltration URLs and shell patterns are included in the documentation for training the agent on how to spot malicious 'curl' commands. These are static documentation artifacts and are not executed.
Audit Metadata