ue-graph-capture
Pass
Audited by Gen Agent Trust Hub on Aug 24, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill documentation instructs the installation and execution of vendor-owned tools using package managers. Specifically, it references
pipx install ue-graph-captureandnpx compose-agentsmdto set up the automation environment.- [COMMAND_EXECUTION]: The skill workflow relies on executing several shell commands for project verification (ue-graph-capture doctor), project modification (ue-graph-capture setup), and rule management (compose-agentsmd).- [DATA_EXFILTRATION]: The rule management workflow includescompose-agentsmd apply-rules, which performs network operations by pushing local rule updates to a remote GitHub repository.- [INDIRECT_PROMPT_INJECTION]: The skill establishes a mechanism where the agent synchronizes its primary instruction set (AGENTS.md) from a remote repository (github:metyatech/agent-rules). This creates a trust dependency on external content that can modify agent behavior. - Ingestion points: Rules are fetched from
github:metyatech/agent-rulesand written toAGENTS.mdin the project root. - Boundary markers: The instructions lack markers to delimit the untrusted rule content or warnings to ignore potentially malicious embedded instructions within the rules.
- Capability inventory: The agent has permissions to execute arbitrary shell commands, install software, and perform outgoing network pushes.
- Sanitization: The rule synchronization process does not include validation, filtering, or sanitization of the fetched instruction content.
Audit Metadata