The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill processes external data (logs) which is a known vector for indirect prompt injection. An attacker who can write to production logs (e.g., via error messages or metadata) could attempt to influence the agent's behavior.
Ingestion points: logs/trades_2025.log, logs/errors.log, and logs/window_*.log.
Boundary markers: Absent. The instructions do not provide delimiters or warnings to ignore instructions found within the logs.
Capability inventory: Read access to logs and execution of the sync_logs and grep tools.
Sanitization: Not present. The skill does not perform validation or escaping of the log content before analysis.

log-analyzer