atlas-crew-tasks
Fail
Audited by Gen Agent Trust Hub on Jul 17, 2026
Risk Level: HIGHCOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to perform privilege escalation by executing
gh auth refresh -h github.com -s admin:org. This requests broad administrative access to a GitHub organization, which poses a significant risk if the agent processes malicious input or is otherwise compromised. - [COMMAND_EXECUTION]: The instructions involve the use of
gh apiandgh api graphqlto modify organizational settings and project structures. These commands can have wide-reaching effects on organization resources beyond simple task management. - [COMMAND_EXECUTION]: The skill references an external command
gh seqfor determining task order. There is no information provided regarding the source, safety, or installation process for this tool, representing an unverifiable dependency in the workflow. - [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8). It ingests untrusted data from GitHub issues (titles and bodies) and uses it within shell commands (e.g.,
gh issue create ... --body "...") and project updates without boundary markers, sanitization, or explicit instructions to ignore embedded commands in the processed data.
Recommendations
- AI detected serious security threats
Audit Metadata