born
Pass
Audited by Gen Agent Trust Hub on Jul 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to perform file system operations, including creating directories ('Islands/peronality', 'memorbs/Dump') and updating workspace configuration files ('CLAUDE.md', 'AGENT.md', 'Headquarter/Core/idenity.md').
- [DATA_EXFILTRATION]: While no external network exfiltration was found, the skill explicitly prompts the user to provide personal and professional details such as name, title, organization, and career goals. This information is stored locally within the vault's core configuration files to establish identity context for the agent.
- [INDIRECT_PROMPT_INJECTION]: The skill implements an interactive interview (Q1-Q4) where user-provided responses are written directly into core instruction files ('CLAUDE.md' and 'AGENT.md'). This establishes a pattern where user input influences future agent behavior. Since this is the primary intended purpose of the initialization skill and occurs during an interactive session, it is considered a functional feature rather than a vulnerability.
Audit Metadata