Show HN: Standing Questions – agent memory that stores questions, not answers (github.com)

A new memory pattern for AI agents, termed "Standing Questions," has been introduced to tackle the challenge of persistent knowledge decay in long-term coding projects. Instead of storing answers that may become outdated as repositories evolve, this approach emphasizes maintaining a set of key questions. At the start of each project session, the AI agent re-derives answers against the current state of the codebase, ensuring that any discrepancies or "deltas" are vocalized. This method not only prevents the silent decay of outdated information but also fosters a culture of continuous reassessment, promoting more accurate and relevant insights. The significance of this pattern lies in its innovative approach to memory management within AI agents, particularly for those that operate over extended periods and across multiple sessions. By treating answers as disposable while preserving questions as durable artifacts, it allows for improved oversight and accountability. The accompanying tools—two JSON schemas, a set of initial questions, and metrics for monitoring stale data—boost the application of this pattern across various operating systems and eliminate dependency issues. This shift towards a question-centric framework may significantly enhance the reliability of AI agents in complex, evolving code environments, mitigating risks associated with relying on static, potentially incorrect information.