The Agent Skills Directory

[COMMAND_EXECUTION]: Uses Bash for environment-aware path construction (date) and log inspection (tail). These are legitimate uses within the context of a log-reading utility and do not involve untrusted input or shell interpolation from external sources.
[DATA_EXFILTRATION]: The skill reads log files that may contain sensitive trading information (account info, symbols). The skill explicitly mitigates exfiltration risk by excluding network-capable tools from the allowed-tools configuration and instructing the agent to filter sensitive data when reporting results.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from MetaTrader 5 log files which could theoretically contain malicious instructions if a third-party indicator is compromised.
Ingestion points: Reads log files located at $MQL5_ROOT/Program Files/MetaTrader 5/MQL5/Logs/*.log.
Boundary markers: No explicit markers (like XML tags or specific delimiters) are used to isolate log content from the agent's instructions.
Capability inventory: Uses Bash (for tail), Read (file access), and Grep (searching).
Sanitization: No automated sanitization of log content is performed before processing, although the documentation advises manual filtering of sensitive output.

log-reader