The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external web searches (LinkedIn, GitHub) and interpolates it into prompts used to generate briefs.
Ingestion points: Research data retrieved via scripts/research_person.js from LinkedIn profiles, GitHub repos, and news sites.
Boundary markers: The prompt template in scripts/generate_brief.js lacks clear delimiters or instructions for the AI to ignore instructions embedded within the research data.
Capability inventory: Access to gmail skill (sending emails), curl (sending to Slack webhooks), and gcalcli (reading calendar data).
Sanitization: The skill performs no escaping or validation of the research data before stringifying it into the prompt.
[DYNAMIC_EXECUTION]: The skill utilizes shell heredocs in scripts/check_calendar.sh to dynamically generate and execute Python scripts at runtime for parsing calendar output.
[COMMAND_EXECUTION]: The system relies on executing multiple CLI tools, including gcalcli for calendar access, gh (GitHub CLI) for profile research, and curl for interacting with Slack webhooks.
[DATA_EXFILTRATION]: The skill is designed to read sensitive information (Google Calendar events and MEMORY.md notes) and transmit it to external endpoints (Gmail and Slack webhooks). While this is the primary stated purpose of the skill, it represents a data transfer of private information to external services.

meeting-brief