The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze external real estate auction notices and property descriptions, which creates a potential surface for indirect prompt injection if the ingested data contains malicious instructions.
Ingestion points: User-provided auction details, property descriptions, and legal notices triggered by keywords defined in SKILL.md.
Boundary markers: The instructions lack specific delimiters or isolation mechanisms to separate untrusted property data from the agent's logic.
Capability inventory: The skill includes a logging script (scripts/governance.py) that writes to a local state directory (~/.local/state/leiloeiro-risco) with restrictive permissions (0o600) and secure file opening flags (O_NOFOLLOW). The skill does not demonstrate network exfiltration or shell execution capabilities.
Sanitization: No evidence of sanitization or filtering of input data is present in the instructions or scripts.
[SAFE]: The skill references established Brazilian government services for environmental risk assessment and legal research.
Evidence: Documented references to cemaden.gov.br and ibge.gov.br for property due diligence and geographic verification.

leiloeiro-risco