cli-agent-delegation

Warn

Audited by Gen Agent Trust Hub on Jul 23, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides command templates for the agent CLI that utilize high-risk flags.
  • The references/cursor-agent-cli.md file documents the use of --force, which is described as enabling auto-run, trusting the workspace, and skipping MCP confirmations.
  • The --trust flag is used to grant workspace trust.
  • [PROMPT_INJECTION]: The delegation mechanism utilizes a "mission packet" to convey instructions to a sub-agent with extensive capabilities.
  • Ingestion points: Context and instructions are aggregated into a mission packet referenced in SKILL.md and references/cursor-agent-cli.md.
  • Boundary markers: The skill defines a structured text format for the mission packet.
  • Capability inventory: The sub-agent has the authority to write files, execute shell commands, and use web tools as granted by the agent CLI flags.
  • Sanitization: The skill instructs developers to avoid eval and use subprocess argument arrays to prevent command injection in the host shell.
  • [EXTERNAL_DOWNLOADS]: The skill references documentation and official resources from the Cursor platform.
  • Evidence: Links to official Cursor documentation and changelogs are provided in references/cursor-agent-cli.md.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 23, 2026, 06:44 AM
Security Audit — agent-trust-hub — cli-agent-delegation