The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process conversation history, which is untrusted input that could contain malicious instructions.
Ingestion points: The active thread and conversation history are processed as primary data sources (SKILL.md).
Boundary markers: The instructions tell the agent to treat non-active context as interpretive only, but no formal data delimiters are used.
Capability inventory: The skill has no capability to execute commands, write files, or access the network. It explicitly sets disable-model-invocation: true in the frontmatter, preventing it from calling other tools.
Sanitization: No specific data sanitization is mentioned, but the output format is limited to a plain-text ledger.
[SAFE]: The skill contains no executable code, persistence mechanisms, or network-enabled operations.
The implementation is restricted to natural language instructions for summarization.
No hardcoded credentials or sensitive file path access were detected.

standup