The Agent Skills Directory

PROMPT_INJECTION (LOW): The skill implements logic for 'Semantic Connection Detection' and 'Insight Generation' which involves the AI reading and analyzing user-generated content from logs and notes. This creates an indirect prompt injection surface where malicious instructions in a log entry could influence the AI's behavior or subsequent insights. \n- Ingestion points: challenge-log.md, today.md, and sessions/session-XXX/notes.md. \n- Boundary markers: Absent; the templates use standard Markdown headers and do not include instructions to ignore embedded commands. \n- Capability inventory: File system read/write for log management and insight generation; optional web search functionality for 'BUILDING' challenge types. \n- Sanitization: No sanitization or escaping of user input is specified before the AI processes the content for summaries and insights. \n- NO_CODE (SAFE): This skill consists entirely of Markdown templates and JSON configuration files. No Python scripts, Node.js code, or shell commands are present, which eliminates the risk of malicious script execution or dependency-related vulnerabilities.

streak