Hierarchical Agent Memory v3.1.0

Agents that run across multiple projects, contacts, and time horizons need memory that scales. Without structure, memory files grow into a single bloated document that triggers lost-in-the-middle degradation. This skill organizes agent memory into a hybrid architecture: topic-based working memory for active projects, and optional time-based archival layers for episodic recall.

Onboarding

When this skill is installed or updated, initiate a setup conversation with the user before applying any memory structure changes. Do not silently create directories or files.

Step 1 — Greet and explain. Tell the user this skill organizes agent memory into a hybrid structure with topic-based working memory and optional time-based archival layers. Keep it to 2-3 sentences.

Step 2 — Ask about preset preference. Present three presets:

• minimal — MEMORY.md routing table plus daily notes. Best for simple setups and single-project agents.
• standard (default) — Adds topic files and contact files. Best for multi-project agents and teams.
• full — Adds weekly, monthly, and yearly auto-distillation layers. Best for long-running agents needing historical recall. If the bundled memory-wiki plugin is enabled, prefer standard — memory-wiki's syntheses/ pages and reports/ dashboards already cover the same ground.

If the user says "just use defaults," apply standard and move on.

Companion skills and plugins: For per-channel session isolation, install agent-session-state. For governance and provenance tracking, install agent-provenance. If the bundled memory-wiki plugin is enabled, it sits beside this skill and is complementary — this skill owns the always-loaded MEMORY.md routing table and memory/topics/ active-project working memory; memory-wiki owns the compiled durable knowledge vault (structured claims, provenance, dashboards). All are independent.