The Agent Skills Directory

[DATA_EXPOSURE]: The skill handles conversation history, which may contain sensitive information. To mitigate risk, it explicitly requires the agent to redact API keys, tokens, passwords, and PII before writing to the filesystem, replacing them with placeholders like [REDACTED].
[COMMAND_EXECUTION]: The instructions include technical guidance on performing arithmetic in shell environments (using the 10# prefix in bash) to correctly increment file numbers without octal parsing errors. This is a logic constraint for the agent's internal operation and does not involve executing untrusted or arbitrary commands.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input from the conversation history and writes it to a file. This creates an indirect prompt injection surface; however, the risk is minimized by the use of structured markdown templates and the absence of any capability that would execute the contents of the generated files.

handoff-create