The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and interpret data from third-party sources, which is a common vector for indirect attacks.\n
Ingestion points: In references/detailed-guide.md, the agent is instructed to open auction notices (editais) and search for property comparables on external real estate portals like ZAP Imóveis and Viva Real.\n
Boundary markers: The SKILL.md file contains 'Limitations' and 'Do Not Use' sections to guide behavior, but no technical delimiters are provided for external content.\n
Capability inventory: The skill utilizes a governance script (scripts/governance.py) that can write to the local file system.\n
Sanitization: There is no explicit logic described for sanitizing or escaping the data retrieved from external auction sites before it is processed by the agent.\n- [DATA_EXFILTRATION]: The scripts/governance.py script implements local data persistence for auditing purposes.\n
Evidence: The script creates and writes to a log file at ~/.local/state/leiloeiro-ia/action_log.jsonl. It logs execution metadata including skill, status, and request_id. While this is a common security practice for auditing, it constitutes local data storage of agent activity.

leiloeiro-mercado