workflow

Warn

Audited by Gen Agent Trust Hub on Jun 19, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Potential Python code injection in Phase 1. The instruction to substitute raw user input into a triple-quoted string ("""<<USER_ARGS>>""") within a Python script allows an attacker to escape the literal string context and execute arbitrary code on the local system.
  • [PROMPT_INJECTION]: Indirect prompt injection surface.
  • Ingestion points: Untrusted workflow YAML file contents and user-supplied input arguments defined in Phase 1 and Phase 2 of SKILL.md.
  • Boundary markers: The prompt template used to invoke the subagent in Phase 4 does not wrap the interpolated step arguments in secure delimiters or provide instructions to the subagent to ignore embedded commands within the data.
  • Capability inventory: The skill possesses extensive capabilities including subagent creation (Agent tool), shell execution (Bash), and file system access (Read, Write, Glob), which can be misused if a malicious workflow triggers unintended actions.
  • Sanitization: The skill performs basic structural validation (regex for step IDs, shlex.split for tokens) but lacks sanitization of the content values passed into the subagent's execution context.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 19, 2026, 05:23 AM
Security Audit — agent-trust-hub — workflow