The Agent Skills Directory

[SAFE]: The skill implements an autonomous research loop that prioritizes security by design. It explicitly defines a 'corpus' as untrusted, ensuring that external code is only read and cited, never built, installed, or executed.
[SAFE]: The skill uses sub-agents for specialized tasks like skepticism and criticism to maintain objectivity and prevent bias. All claims must be backed by cited evidence from a 'source ledger' which ranks official documentation and peer-reviewed research over less reliable sources.
[SAFE]: It handles external data ingestion (web searches, corpus reading) with a defined 'read-only' constraint, minimizing the risk of accidental execution or system compromise.
[SAFE]: The 'Wayfinder' seed handoff mechanism ensures a clear separation of concerns between decision-making and implementation, preventing the agent from autonomously transitioning from research to potentially dangerous file-writing or execution tasks without a new context.

storm