The Agent Skills Directory

[DATA_EXFILTRATION]: The skill aggregates sensitive business information, including email history, meeting notes, and attendee research, and exfiltrates it to external third-party services (WhatsApp and Slack) as part of its briefing process.
[COMMAND_EXECUTION]: Spawns shell processes via curl to interact with external APIs, specifically using Slack's file upload feature with bot tokens and parameters derived from local configuration files.
[PERSISTENCE]: Automatically manages its own persistence by using openclaw cron to schedule one-shot and recurring tasks (reminders and post-meeting processors) that trigger future agent activity.
[INDIRECT_PROMPT_INJECTION]: Significant vulnerability surface identified. The skill ingests unvetted content from Gmail threads, LinkedIn profiles, and company websites. Malicious instructions hidden in these external sources could be ingested into the agent's context, potentially influencing the summaries, focus areas, or the parameters of the automated follow-up tasks.
[DYNAMIC_EXECUTION]: Dynamically generates natural language instructions ("tasks") based on runtime meeting data and injects them into the cron system for future execution by the AI agent.

meeting-prep