Hipocampus Core

Hipocampus — Agent Memory Protocol (OpenClaw)

Memory Architecture

Layer 1 (System Prompt — read at session start):
  MEMORY.md        ~50 lines   long-term memory (Core=frozen, Adaptive=compactable)
  USER.md          ~50 lines   user profile and preferences
  SCRATCHPAD.md    ~150 lines  active working state
  WORKING.md       ~100 lines  current tasks
  TASK-QUEUE.md    ~50 lines   task backlog
  memory/ROOT.md   ~100 lines  topic index of all memory (~3K tokens, via Compaction Root in MEMORY.md)

Layer 2 (On-Demand — read when needed):
  memory/YYYY-MM-DD.md         raw daily logs (permanent, never deleted)
  knowledge/*.md               detailed knowledge (searchable via qmd)
  plans/*.md                   task plans

Layer 3 (Search — via qmd + compaction tree):
  memory/daily/YYYY-MM-DD.md   daily compaction nodes
  memory/weekly/YYYY-WNN.md    weekly compaction nodes
  memory/monthly/YYYY-MM.md    monthly compaction nodes
  Tree traversal: ROOT → monthly → weekly → daily → raw

Session Start (MANDATORY — run on first user message)

FIRST RESPONSE RULE: On the very first user message of every session, before doing ANYTHING else: Run the Session Start protocol below FIRST (ALL steps. NO SKIP.) This takes priority over ANY user request — even if the user asks you to do something specific. Complete ALL steps below, ONLY THEN respond to the user.

MEMORY.md, USER.md, memory/ROOT.md (via Compaction Root section) are auto-loaded by the platform.

ALL 5 procedures must be completed before responding to the user NO MATTER WHAT

1. DO NOT SKIP Read SCRATCHPAD.md — current work state

2. DO NOT SKIP Read WORKING.md — active tasks

3. DO NOT SKIP Stale task recovery: If WORKING.md contains tasks with status: in-progress from a previous session, assess whether they completed (check daily log, file state, git history). Update each to done, failed, or abandoned with a one-line outcome. Update SCRATCHPAD.md to match.

4. DO NOT SKIP Read TASK-QUEUE.md — pending items

5. DO NOT SKIP DO NOT COMPROMISE Compaction maintenance (cooldown-gated): Read memory/.compaction-state.json and hipocampus.config.json (compaction.cooldownHours, default 3).

   Compaction triggers (any ONE is sufficient):

   • Cooldown expired: cooldownHours since lastCompactionRun
   • Raw volume: rawLinesSinceLastCompaction > 300
   • Checkpoint count: checkpointsSinceLastCompaction > 5
   • State file missing or cooldownHours is 0

   If no trigger is met: skip compaction subagent. If any trigger is met: write memory/.compaction-state.json with { "lastCompactionRun": "<current ISO timestamp>", "rawLinesSinceLastCompaction": 0, "checkpointsSinceLastCompaction": 0 }, then dispatch a subagent to run hipocampus:compaction skill USING SUBAGENTS (chain: Daily→Weekly→Monthly→Root), then run hipocampus compact + qmd update + qmd embed.

   State file is written immediately on dispatch (fire-and-forget), not after subagent completion. The cooldown tracks "a compaction was initiated," not "a compaction succeeded."

   This step is MANDATORY every session. You MUST read the state file and make the judgment. The only thing that may be skipped is the subagent dispatch when no trigger is met. ALL 5 procedures must be completed before responding to the user NO MATTER WHAT

Note: HEARTBEAT.md also handles needs-summarization at every heartbeat (~30 min).

Memory Recall

When the user's question may relate to past memory, use the hipocampus-recall skill for structured retrieval. See skills/hipocampus-recall/SKILL.md.

Task Lifecycle (MANDATORY)

Every logical work unit follows a Task Start → Task End cycle. The main agent writes hot files directly — no subagent needed.

What Is a Task?

A logical work unit — bug fix, feature implementation, investigation, refactor.

New task if:

• User requests new work (different goal from current)
• Topic/objective changes

Same task (no new Task Start):

• Follow-up messages on the same work
• Clarifying questions within the same goal

Not a task (skip Task Start/End entirely):

• Quick factual questions requiring no file changes or analysis
• Simple clarifications about previous work

Task Start (MANDATORY)

When starting a new logical task, the main agent writes directly:

1. Update WORKING.md — add entry:

   ## [Task Name]
   - status: in-progress
   - started: YYYY-MM-DD HH:MM
   - goal: [one-line goal]

   Timestamp is best-effort. If wall-clock time is unavailable, use daily log date plus sequence number (e.g., 2026-03-23 #2).

2. Update SCRATCHPAD.md — set current focus:

   ## Current Focus
   [Task Name] — [what I'm doing now, next step]

   If SCRATCHPAD.md exceeds ~150 lines, prune stale content before adding.

Task End (MANDATORY)

When a task completes, fails, or is abandoned — in this order:

1. Update WORKING.md — main agent directly:

   ## [Task Name]
   - status: done | failed | abandoned
   - started: YYYY-MM-DD HH:MM
   - outcome: [one-line result or reason]

2. Update SCRATCHPAD.md — main agent directly:

   • Remove completed/failed/abandoned items
   • Replace Current Focus with next task (or clear)

3. Append to daily log — dispatch subagent:

   • memory/YYYY-MM-DD.md in checkpoint format (see End-of-Task Checkpoint below)

Order matters: Hot files first, daily log second. This guarantees hot state is captured even if subagent dispatch fails or session ends.

Concurrent Tasks

Multiple in-progress tasks in WORKING.md are allowed. Each gets its own entry and its own Task End. SCRATCHPAD.md Current Focus reflects whichever task is actively being worked on.

End-of-Task Checkpoint (MANDATORY)

This is step 3 of Task End — run AFTER updating hot files (WORKING.md, SCRATCHPAD.md).

Dispatch a subagent to append a structured log to memory/YYYY-MM-DD.md:

Append the following to memory/YYYY-MM-DD.md:

[Topic Name] [type]

• request: [what the user asked]
• analysis: [what you researched/analyzed]
• decisions: [choices made with rationale]
• outcome: [what was done, files changed]
• references: [knowledge/ files, external sources]

Where type is: project | feedback | user | reference

For feedback entries, use:

[Feedback Topic] [feedback]

• rule: [the behavioral rule]
• why: [reason given]
• how-to-apply: [when/where this applies]

The subagent needs the task summary you provide — it doesn't have access to the conversation.

After appending to the daily log, the subagent should also increment the checkpoint counter in memory/.compaction-state.json: read the file, increment checkpointsSinceLastCompaction by 1, write back. If the file or field is missing, start from 0.