The Agent Skills Directory

[SAFE]: The skill is designed to manage session state by summarizing conversations into local markdown files within the project's .codex/handoffs/ directory.
[DATA_EXPOSURE_AND_EXFILTRATION]: The instructions include a mandatory redaction step for sensitive information, such as API keys, passwords, or personally identifiable information (PII), which mitigates the risk of accidental secret exposure in handoff documents.
[INDIRECT_PROMPT_INJECTION]: Although the skill processes untrusted conversation data and user arguments, the instruction to redact sensitive data and the limited scope of the file-writing capability minimize the potential impact of adversarial content embedded in the conversation history.

handoff