chief-of-staff
Pass
Audited by Gen Agent Trust Hub on Jul 15, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill uses a custom invocation syntax
[INVOKE:role|question]to orchestrate 28 other specialized skills. This is a standard pattern for multi-agent systems and is governed by strict loop prevention rules (maximum depth of 2 and circular blocking) to ensure stability. - [DATA_EXPOSURE]: The skill maintains a persistent record of decisions by reading from and writing to
~/.claude/decision-log.md. This behavior is transparently documented as a 'Decision Log' feature for tracking business outcomes and providing automated review reminders. - [PROMPT_INJECTION]: While the skill processes untrusted user input and outputs from other agents, it employs a 'Synthesis Framework' that explicitly requires identifying themes and surface conflicts. This structural approach to processing information reduces the risk of the agent accidentally executing instructions hidden within the data it analyzes.
Audit Metadata