Consolidate an agent's raw events into durable memory facts.
npx skills add https://github.com/AbyssCN/xihe --skill dream-abysscn
Please help me install this Agent Skill. Skill: dream Source: https://github.com/AbyssCN/xihe/tree/main/skills/dream Command: npx skills add https://github.com/AbyssCN/xihe --skill dream-abysscn
This Skill manually triggers Dream consolidation, distilling an agent's raw events into durable memory facts with safeguards.
Run the /dream command to manually consolidate memory.
High-intent search queries and answers about installing and using this skill.
To manually consolidate agent memory from raw events, trigger the consolidation process using commands like /dream or "consolidate memory". This distills raw agent events into durable memory facts, ensuring reliability and adherence to a budget cap.
Memory consolidation for agent events is the process of distilling raw activity logs into durable memory facts. You need it after a burst of activity or before a long break to capture lessons and ensure the next session starts with distilled knowledge.
Yes, there is a budget cap of approximately 10,000 tokens when distilling facts from agent events. This limitation ensures all consolidation layers run reliably within a constrained memory budget without exceeding resource limits.
To trigger memory consolidation and distill facts, use trigger words such as /dream, "dream run", "consolidate memory", or "整理记忆". Running these commands manually activates the fact distillation process for your agent's raw events.
Fact distillation works reliably for long-term agent memory by manually triggering the consolidation of raw events into durable facts. It includes safeguards and a budget cap to ensure the distilled knowledge remains accurate and resource-efficient.
brucesongsAggregates raw agent lifecycle events into structured chronicles with CHRONICLE.md indexing and MEMORY.md distillation.
ascorbicExtract actions, facts, and decisions from conversations into structured JSON.
diptanshuagrawalAutomate scheduled Slack data ingestion into a normalized event schema.
saolalabStore long-term facts in MEMORY.md and searchable HISTORY.md event logs.
