foundation-meeting-agenda
Pass
Audited by Gen Agent Trust Hub on Apr 27, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill instructions define clear, task-oriented logic for generating meeting agendas. There are no attempts to override safety guidelines, extract system prompts, or bypass agent constraints. The 'Zero-friction execution' mode is a behavioral configuration to reduce user interrogation, not a bypass mechanism.
- [DATA_EXFILTRATION]: The skill does not perform network operations to external domains. The URLs provided in the EXAMPLE.md file (e.g., notion.example, meet.example) are placeholders for illustrative purposes and do not point to real or malicious infrastructure. No access to sensitive local file paths (e.g., .ssh, .aws) was detected.
- [OBFUSCATION]: A manual and automated scan of the instruction files (SKILL.md, EXAMPLE.md, TEMPLATE.md) revealed no hidden content, base64-encoded commands, homoglyphs, or zero-width character manipulation.
- [REMOTE_CODE_EXECUTION]: The skill does not download or execute external scripts, packages, or binaries. It relies entirely on the agent's internal text processing and template rendering capabilities.
- [COMMAND_EXECUTION]: There are no shell commands, subprocess calls, or dynamic execution patterns like 'eval' or 'exec' present in the instruction set. The skill does not use the dynamic context injection syntax ('!command') at skill load time.
- [INDIRECT_PROMPT_INJECTION]: While the skill ingests external data (user topics and @file references), its capability inventory is restricted to text generation. It lacks dangerous write-access or network capabilities that would allow an indirect injection to escalate into a more severe compromise.
Audit Metadata