The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external CSV files exported from Ghostwriter. This creates a potential attack surface where malicious instructions could be embedded in the log entries.
Ingestion points: .csv files located in the input/gw_oplog/ directory.
Boundary markers: The instructions do not define any boundary markers or delimiters to separate untrusted data from instructions during processing.
Capability inventory: The skill performs data normalization and mapping, writing the results to output/gw_entries.json. The instructions do not specify any dangerous tool invocations (like shell execution) based on the input, but the structured output could be consumed by downstream processes.
Sanitization: The skill includes validation for required columns and timestamp normalization, but lacks sanitization or filtering for malicious prompt content within free-text fields like 'description', 'command', or 'details'.

timeline-ghostwriter