claude-handoff

Pass

Audited by Gen Agent Trust Hub on Jul 27, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructions direct the agent to invoke the claude CLI via the shell to spawn background processes using the command template claude --bg --name "<descriptive name>" "<handoff summary>". This involves executing shell commands with strings generated dynamically from conversation data.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it incorporates untrusted data from the conversation history and user-supplied arguments into a summary that serves as the system prompt for a new background agent.
  • Ingestion points: The entire conversation history and user-provided arguments are used as context for the summary (SKILL.md).
  • Boundary markers: No explicit delimiters or boundary instructions are used to isolate untrusted data within the command template or the resulting agent prompt.
  • Capability inventory: The skill enables shell command execution through the platform's CLI to initiate background tasks (SKILL.md).
  • Sanitization: The instructions provide a manual redaction requirement for sensitive data (API keys, passwords), but do not implement technical sanitization for shell injection or prompt injection safety.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 27, 2026, 10:24 AM
Security Audit — agent-trust-hub — claude-handoff