Openclaw Skills for Dawn Memory Architecture v7.1 creates a local, layered memory system for agents with truth hierarchy, WAL safety, and semantic search.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install dawn-memory-arch-v7
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install dawn-memory-arch-v7 using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Dawn Memory Architecture v7.1?
Dawn Memory Architecture v7.1 is an Openclaw Skills package for building durable long-term agent memory with a strictly local storage model. It combines a central MEMORY.md index, structured JSON facts, session and learning logs, and a local vector database so an agent can retrieve, update, and promote knowledge without relying on external APIs.
The core value is disciplined memory management: keep high-trust facts in the right layer, preserve history separately from decision-making, and use write-ahead logging plus automated promotion rules to prevent data loss and memory bloat. Openclaw Skills users get a reproducible architecture for context management, truth hierarchy, and offline semantic recall.
Dawn Memory Architecture v7.1 Use Cases
- Building long-term memory for AI coding or trading agents that must stay consistent across sessions.
- Maintaining a local-only knowledge base with no external API dependency.
- Managing structured facts such as identity, preferences, strategies, lessons, and profile data.
- Preserving session history separately from decision inputs for safer agent behavior.
- Automatically promoting high-value insights from logs into a central index.
- Using semantic search to retrieve relevant memories from past work quickly.
- Preventing context overflow with working-buffer summaries and threshold-based capture.
- Supporting post-mortem analysis, correction tracking, and durable state recovery.
How Dawn Memory Architecture v7.1 Works
- On startup, the agent reads
MEMORY.mdas the central navigation index and loads the current system status, key conclusions, and promoted memories. - Structured truth is stored in
memory/core/*.jsonfor stable facts such as identity, preferences, strategies, lessons, and profile data. - Session history is written to
memory/sessions/, while daily notes, learnings, and error logs are stored in separate folders to keep decision memory clean. - When corrections, preference changes, or decisions occur, the WAL protocol writes to
session-state.jsonfirst so changes are not lost on crash. - If context usage crosses the 60% threshold, the working buffer captures a compact summary in
working-buffer.mdto preserve the active thread. - High-value items are auto-promoted from lower-tier logs into
MEMORY.mdbased on score, repetition, or explicit user request. - A local vector database built with LanceDB and all-MiniLM-L6-v2 enables offline semantic search across sessions and notes.
- Maintenance scripts validate schemas, sync boot state, update trade snapshots, and run daily analysis tasks on a schedule.
Dawn Memory Architecture v7.1 Setup
- Create the workspace structure exactly as specified, including
MEMORY.md,memory/core/,memory/sessions/,memory/daily/,memory/learnings/,memory/archive/,scripts/memory/,scripts/analysis/, andstate.db. - Initialize the central index and keep it under the documented hard limit.
- Create the WAL cache and operational checklist files.
- Deploy the local vector database stack with LanceDB and the embedded MiniLM model.
- Register the curation, promotion, and validation rules in your boot workflow.
- Add the maintenance scripts and wire them into startup and trading-day triggers.
mkdir -p workspace/memory/core workspace/memory/sessions workspace/memory/daily workspace/memory/learnings workspace/memory/archive workspace/scripts/memory workspace/scripts/analysis
touch workspace/MEMORY.md workspace/session-state.json workspace/HEARTBEAT.md workspace/state.db
# Example bootstrap tasks
python scripts/memory/dawn_boot_sync.py
python scripts/memory/dawn_schema_validate.py
python scripts/memory/dawn_state_db_upgrade.py
python scripts/memory/dawn_money_flow.py
python scripts/memory/dawn_factor_scoring.py
- Configure Openclaw Skills workflows so
MEMORY.mdis always read on startup,session-state.jsonis written before replies when needed, and context summaries trigger automatically at 60% usage.
Dawn Memory Architecture v7.1 Data Schema & Taxonomy
Storage layers
| Layer | Path | Purpose | Notes |
|---|---|---|---|
| P0 | state.db |
Single source of truth for live positions and portfolios | Updated automatically after each trade |
| P1 | memory/core/*.json |
Structured cache for stable facts | Manual + auto validation |
| P2 | memory/sessions/*.md |
Session history | Not used for decision-making |
| P3 | memory/learnings/*.md |
Post-mortem lessons and error notes | Recorded after the fact |
Core files
MEMORY.md: central index, max 200 lines or 25KB.identity.json: identity and persona facts.lessons.json: learned rules and outcomes.preferences.json: user preferences and constraints.profile.json: expanded profile system introduced in v7.1.strategies.json: strategic guidance and decision rules.session-state.json: WAL cache for crash-safe writes.working-buffer.md: context fuse summary when usage is high.HEARTBEAT.md: periodic task checklist.
Metadata taxonomy
core: high-trust structured facts.sessions: chronological history, not authoritative.daily: day-by-day operational notes.learnings: retrospective improvements and mistakes.archive: packaged historical snapshots.scripts/memory: operational maintenance jobs.scripts/analysis: analysis and scoring utilities.
Truth hierarchy
- P0: live data and real-time state.
- P1: validated structured memory.
- P2: session logs for reference only.
- P3: learning notes for post-mortem use only.
Openclaw Skills implementation should treat lower layers as support data, never as the authoritative source for live decisions.
Dawn Memory Architecture v7.1 Advanced Features
- Local-only semantic search with LanceDB and all-MiniLM-L6-v2 for offline retrieval.
- WAL-first writes that reduce data loss during crashes or interrupted sessions.
- Automatic memory promotion based on score, repetition, and explicit user intent.
- Context-fuse buffering that captures summaries before the model overflows.
- Strict deduplication and update-over-insert curation to prevent memory drift.
- P0-P3 truth hierarchy that cleanly separates live decision data from historical notes.
- Schema validation hooks for
core/*.jsonchanges. - Startup and trading-day automation through dedicated maintenance scripts.
- Zero external API dependency, making the full stack portable and self-hosted.
- Designed for Openclaw Skills workflows that need persistent, auditable agent memory.
SKILL.md
Loading
Related Openclaw Skills
Project Delivery Engine
Project Delivery Engine helps AI agents reliably manage long-running projects with persistent state, structured handoffs, and independent review.
OpenClaw Deal Scout
Openclaw Deal Scout automates Gmail deal triage, CRM logging, alerts, and approved replies through a local MCP gateway.
USCIS Case Status
A headless command-line skill that checks USCIS case status by receipt number using Selenium and undetected-chromedriver.
DHX Electronic Components Sourcing Agent
Openclaw Skills for DHX Tech turns BOMs, part numbers, and sourcing requests into real-time inventory checks, cross-references, and procurement-ready recommendations.
ai-content-creator-assistant
ai-content-creator-assistant generates multiple platform-adapted content drafts from one topic, then delivers them through a compliant, payment-verified Openclaw Skills workflow.
Requirement Processor
Requirement Processor standardizes how Openclaw Skills turns unclear feature requests into validated, phased implementation plans and reusable project documentation.
Featured*
