The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill processes session logs and historical daily logs which contain data from past interactions, creating a surface for indirect prompt injection.
Ingestion points: scripts/extract_session.py reads session history from .jsonl files, and scripts/search_memory.py reads existing markdown log files.
Boundary markers: Absent. The summaries and search results generated by the scripts do not include explicit delimiters or instructions for the agent to ignore potential commands embedded within the historical data.
Capability inventory: The skill includes scripts for reading files (scripts/search_memory.py), writing/appending to files (scripts/daily_log.py), and summarizing session data (scripts/extract_session.py).
Sanitization: Absent. The scripts read and display or write content directly without filtering or escaping techniques to prevent the LLM from executing instructions found in the text.

memory-curator