The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to send artifact content (such as code and design documents) to external review services via the external_review and consensus_query MCP tools. This behavior is documented as a core feature for enabling multi-model consensus and external security perspectives during the quality gate process.
[COMMAND_EXECUTION]: During the pre-flight phase for code artifacts, the skill executes standard ecosystem audit tools including npm audit, cargo audit, and pip-audit. These tools perform network operations to fetch vulnerability data from official registries to check project dependencies for known security issues.
[PROMPT_INJECTION]: The skill processes untrusted user-supplied artifacts that may contain indirect prompt injection attempts. It mitigates this risk by using specialized subagents (red-teamers, judges, fixers) with narrow, focused prompts and strictly controlled context boundaries, preventing the orchestrator from directly executing instructions embedded within the reviewed artifacts.

quality-gate