Security audit
Security checks for vulnerabilities and agentic risk
This skill is a disclosed multi-agent planning workflow that writes task notes locally, with no evidence of hidden code, exfiltration, persistence, or privilege escalation.
Install this only if you want a formal multi-agent workflow for larger tasks. Expect it to create local session notes in your current project and potentially start multiple sub-agent tasks; confirm the intended scope before letting it execute work, and clarify whether outputs should go under court-session or .court-session.
Several trigger phrases such as common conversational requests overlap with everyday language, making accidental invocation plausible. Because this skill orchestrates structured planning/review/execution and writes artifacts to disk, false activations can lead to unintended task creation and downstream actions that the user did not explicitly authorize.
The trigger scope is broad enough to activate on many ordinary task-oriented requests, causing the agent to enter a powerful multi-step workflow when the user may not have intended it. In a skill that coordinates multiple agents, planning, execution, and file creation, over-triggering increases the chance of unnecessary tool use, unintended side effects, and expanded attack surface from benign conversation.
The documentation first says outputs must go to court-session/{YYYYMMDD-任务名}/, but the concrete example and warning below require .court-session/. These are mutually inconsistent instructions about where artifacts should be written, creating intent-code/documentation divergence within the skill definition.
