Skill

Agent Memory Hierarchy

Structure an OpenClaw agent's memory like a computer — using a cache hierarchy (hot/warm/cold), a YAML fact store for directly addressable data, a lookup ind...

Verified: 2026-05-15 (clawhub-ingest-2026-05-15)

When to use Agent Memory Hierarchy

Choose if

You're rebuilding an OpenClaw or similar agent's memory file and want a structured, three-tier (hot/warm/cold) layout — YAML for directly-addressable facts, a lookup index, and prose only where it earns its keep. The SKILL.md targets a sub-80-line / ~1000-token hot cache, which is the right shape if your agent keeps blowing its context window on memory reads.

Avoid if

Your agent has a small, stable memory that comfortably fits a single prose file — adopting the hierarchy adds discipline overhead (one-line YAML values, no multi-line prose in facts.yaml, separate lookup index) that's only worth it once your memory has grown unwieldy. Also skip if your runtime can't parse YAML or manage multiple memory files.

Risk Flags

  • LOW scope Methodology skill — the SKILL.md provides structure and rules ("one-line values only — no multi-line prose"; "if a value needs explanation, it belongs in MEMORY.md prose instead") but does not ship a migration tool. The agent must do the rewriting work itself.

Cost

Type: Unknown

Distribution

ClawHub
agent-memory-hierarchy
License
MIT-0