Managing context in long-run agentic applications

Slack keeps long security investigations moving by carrying forward a curated record between agent invocations. Its context-management account describes a coordinator's journal of decisions and questions, evidence reviews, and a consolidated event timeline. These replace the accumulated conversation history between invocations, limiting how much old material each new task must process.

The record also preserves a way to challenge conclusions. A reviewing agent can inspect the underlying tool arguments and results, then identify unsupported findings or gaps in the timeline. This extends the earlier investigation design: continuity comes from maintained evidence and decisions, while credibility scores remain the reviewer's judgments.