tsg-flow-validator

Warn

Audited by Gen Agent Trust Hub on Nov 17, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to extract a gate command from the frontmatter of task files and execute it directly. This allows the content of the files being audited to dictate arbitrary shell commands that the agent will run.
  • [DYNAMIC_EXECUTION]: The skill dynamically identifies and executes build, test, lint, and formatting steps by inspecting CI workflows and project scripts at runtime. This involves running unverified code discovered within the user's project directory.
  • [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to malicious instructions embedded in project files, such as task metadata and CI definitions, which can control the agent's behavior and command execution.
  • Ingestion points: Task frontmatter fields (gate, task_kind, gate_expect), CI workflows, and project scripts.
  • Boundary markers: No delimiters or warnings are specified to prevent the agent from obeying instructions embedded in the data it processes.
  • Capability inventory: Shell command execution, git worktree management, and file system writes for reporting.
  • Sanitization: Sanitization is explicitly omitted; the instructions require the agent to "execute the declared gate, without altering it."
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 17, 2026, 10:29 PM