The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to launch background processes by executing the host environment's CLI tool.
Evidence: Instructions in SKILL.md direct the agent to run the command claude --bg --name "<descriptive name>" "<handoff summary>".
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface where untrusted user input could influence the command executed in the shell.
Ingestion points: User-provided arguments and conversation history are ingested in SKILL.md.
Boundary markers: No specific shell-escaping markers or delimiters are defined for the summary string within the command template.
Capability inventory: The skill uses shell execution capabilities to launch the claude CLI (SKILL.md).
Sanitization: The instructions explicitly mandate the redaction of sensitive information like API keys and passwords, though it relies on the model's instruction following rather than programmatic escaping.
[DATA_EXPOSURE]: The skill handles potentially sensitive conversation data by passing it as a prompt to a new agent instance.
Evidence: SKILL.md contains a specific safety instruction: "Redact any sensitive information, such as API keys, passwords, or personally identifiable information, since the summary becomes the agent's prompt."

claude-handoff