skills/uipath/skills/uipath-agents/Gen Agent Trust Hub

uipath-agents

Pass

Audited by Gen Agent Trust Hub on Oct 26, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFEREMOTE_CODE_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill provides instructions and patterns for building agents that ingest and process untrusted data from external files, which is a known vector for indirect prompt injection.
  • Ingestion points: Data enters the agent's reasoning context through the analyze-attachments, deep-rag, and batch-transform tools which process user-provided PDF, TXT, and CSV files (found in references/lowcode/capabilities/built-in-tools/).
  • Boundary markers: Example prompts provided in the documentation (e.g., in references/lowcode/capabilities/built-in-tools/analyze-attachments.md) lack explicit delimiters or instructions to the LLM to ignore potentially malicious commands embedded within those files.
  • Capability inventory: The agents being built have access to powerful tools including shell execution (Bash), file system modification (Write, Edit), and the ability to trigger external RPA processes or create human-in-the-loop tasks via the UiPath platform.
  • Sanitization: The documentation does not provide specific guidance on sanitizing or filtering content extracted from external files before it is processed by the language model.
  • [CREDENTIALS_UNSAFE]: The skill includes shell script examples that programmatically read authentication data from a sensitive local file path associated with the vendor's CLI tool.
  • Evidence: In references/lowcode/capabilities/process/solution-files.md and references/lowcode/capabilities/escalation/escalation.md, the documentation provides a bash -c command that explicitly sources ~/.uipath/.auth to extract UIPATH_ACCESS_TOKEN. While this is used for legitimate product functionality, it involves programmatic access to sensitive credential stores.
  • [REMOTE_CODE_EXECUTION]: The skill's documentation includes instructions to download a script from a remote server and pipe it directly into a shell for execution.
  • Evidence: In SKILL.md, references/coded/quickstart.md, and references/lowcode/project-lifecycle.md, users are instructed to install the uipath-cli using the command curl -fsSL https://download.uipath.com/uipath-cli/install.sh | bash. Although this originates from the official vendor domain, it follows a high-risk remote execution pattern.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 26, 2026, 09:34 PM