The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various Bash commands (find, grep) and specialized CLI tools (uip) to perform its auditing functions. These are used for file discovery and validation within the project scope.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection as it is designed to ingest and process untrusted external data, including Process Design Documents (PDDs) in various formats and project source code.
Ingestion points: Reads user-provided PDDs (PDF, Markdown, Word) and project source files (.xaml, .cs, .py, .json).
Capability inventory: Executes shell commands for searching and uses uip CLI for validation; generates detailed reports.
Boundary markers: The skill instructions provide specific schemas and catalogs for findings, but do not implement strict delimiters or 'ignore' instructions for the content of the analyzed files.
Sanitization: No explicit sanitization of the untrusted content before processing is mentioned, although the agent is instructed to use authoritative CLI findings and predefined rule IDs.
[PROMPT_INJECTION]: The skill's reference documentation (references/agents/agents-lowcode-rules.md) contains strings associated with jailbreak attempts (e.g., 'DAN', 'ignore previous'). These are explicitly defined as criteria for a security scanner to detect in target projects and do not represent a malicious attempt by the skill itself to subvert agent instructions.

uipath-review