uipath-human-in-the-loop

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFECREDENTIALS_UNSAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill instructs the agent to access the sensitive local authentication file at ~/.uipath/.auth to retrieve the UIPATH_ACCESS_TOKEN and other session-related metadata.
  • [COMMAND_EXECUTION]: The skill performs several shell-based operations including binary path resolution using npm root, project scaffolding using rsync, and workflow validation via the uip CLI. It also uses node -e to dynamically generate UUIDs for project identifiers.
  • [DATA_EXFILTRATION]: The skill uses retrieved bearer tokens to perform authenticated network requests to the UiPath Orchestrator API (cloud.uipath.com). While directed at official vendor infrastructure, this pattern involves the automated handling and transmission of sensitive session data.
  • [INDIRECT_PROMPT_INJECTION]: The skill interprets various external project files (such as .flow, caseplan.json, and action-schema.json) to design and insert new workflow nodes, creating a potential surface for indirect injection.
  • Ingestion points: Project configuration and schema files containing external data.
  • Boundary markers: No explicit use of delimiters or 'ignore' instructions for processed data was found.
  • Capability inventory: File writes, network transmission of tokens, and shell command execution.
  • Sanitization: The skill lacks explicit instructions for sanitizing or validating data read from external project files before using it to generate new workflow components.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 06:04 PM
Security Audit — agent-trust-hub — uipath-human-in-the-loop