The Agent Skills Directory

[PROMPT_INJECTION]: The skill utilizes structured instructions to guide the agent through a multi-stage verification process. It enforces 'hard rules' for evidence gathering, such as requiring line-level verification and resolving constant values, which acts as a safeguard against taking potentially malicious or incorrect instructions from processed data at face value.
[DATA_EXFILTRATION]: Access to the file system is properly scoped in the configuration to specific project-related directories (docs, src, lib, app). This restriction ensures the agent only examines relevant code and documentation, preventing unauthorized access to sensitive system configuration files or user credentials.
[COMMAND_EXECUTION]: The skill uses the Bash tool for local logging of its results to .great_cto/triage-log.jsonl. This activity is transparently documented and serves as an audit trail for the triage decisions, with no evidence of malicious command injection or unauthorized network activity.

skeptical-triage