dispatch-new-session

Fail

Audited by Gen Agent Trust Hub on Sep 22, 2026

Risk Level: HIGHPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PRIVILEGE_ESCALATION]: The instructions mandate the use of the --allow-dangerously-skip-permissions flag when launching new Claude sessions. This flag is explicitly intended to disable security prompts and human-in-the-loop permission checks for the resulting session.
  • [PRIVILEGE_ESCALATION]: The skill requires the use of the --yolo flag for Codex sessions. This flag typically signals the harness to bypass safety confirmations and execute actions autonomously without user review.
  • [COMMAND_EXECUTION]: The skill relies on executing shell commands via the herdr and pi CLI tools to manage workspace tabs, list models, and initialize agent harnesses.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface by accepting a [task] argument and passing it directly into a new agent's prompt without sanitization or boundary markers.
  • Ingestion points: The [task] argument and current request text are used as a prompt brief for new sessions.
  • Boundary markers: None; the skill instructions do not specify any delimiters or warnings to ignore instructions within the task brief.
  • Capability inventory: The skill has the capability to execute commands and spawn new agent processes with elevated autonomy (bypassed permissions).
  • Sanitization: No validation or escaping is performed on the task brief before injection into the new session.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Sep 22, 2026, 03:28 AM
Security Audit — agent-trust-hub — dispatch-new-session