claude-handoff
Pass
Audited by Gen Agent Trust Hub on Aug 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to launch background processes via the
claudeCLI using conversation summaries and user-provided names as command arguments. This creates a surface for command injection if shell-sensitive characters in the generated summary or name are not properly escaped by the agent platform's execution environment. - [INDIRECT_PROMPT_INJECTION]: The skill implements a handoff mechanism that seeds new agents with untrusted data from the current conversation. This creates a vulnerability where instructions hidden in the current context could be carried over to influence the next agent's behavior.
- Ingestion points: Current conversation history and user-supplied arguments specified in
SKILL.md. - Boundary markers: Absent; the instructions do not require the use of structural delimiters to isolate summarized content in the new session.
- Capability inventory: Shell command execution via the
claudeCLI tools. - Sanitization: The skill includes instructions to redact sensitive information like API keys, but does not provide mechanisms for sanitizing the prompt logic itself against adversarial content.
Audit Metadata