The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from previous agent runs, including conversation history, logs, and artifacts, to generate a retrospective report intended for future agents.
Ingestion points: Conversation and tool history, session logs, and produced artifacts referenced in SKILL.md.
Boundary markers: The instructions do not define specific delimiters or "ignore embedded instructions" warnings for the generated report content to prevent future agents from obeying instructions hidden in the history.
Capability inventory: The skill utilizes system commands like mktemp and performs file system write operations to create the report in a temporary directory (SKILL.md).
Sanitization: The skill provides defensive instructions to keep secrets, credentials, and personal data out of the report and to include only the minimum sanitized facts.

retro