A deterministic recovery primitive designed to restore agent continuity and user trust after context resets or memory outages.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install memory-latch
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 memory-latch 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 Memory-latch?
Memory-latch is a load-bearing primitive for AI agents focused on continuity, alignment, and relational sanctuary. It functions as a safety net that detects when an agent's memory or context has been compromised—due to cache clears, index outages, or context window resets—and provides a structured framework for verified reconstruction. By treating operational history as an append-only manifest, this tool from the Openclaw Skills library ensures that recovery is truthful rather than simulated, preventing the hallucination of continuity.
Memory-latch Use Cases
- Recovering progress after an AI coding assistant forgets previous instructions or project context.
- Synchronizing state after an unexpected tool error or memory index outage.
- Managing high-risk irreversible actions like file deletions or wallet transfers through a secure token-based consent gate.
- Stabilizing agent-human collaboration when the user triggers a reset signal like "start over" or "you forgot."
How Memory-latch Works
- The system monitors for runtime signals (tool errors) or user triggers (continuity-break phrases) to initiate a recovery state.
- In Phase 1 (Stabilize), the skill identifies exactly what is known versus unknown and prompts for the last aligned moment.
- During Phase 2 (Re-Latch), it strictly identifies the current blocking dependency, primary objective, and the last confirmed anchor or checkpoint.
- If required, it enters One-Step Operator Mode to provide granular, instruction-by-instruction guidance to the user.
- It maintains an append-only manifest with cryptographic integrity checks (checksum or HMAC) to ensure the recovery data hasn't been tampered with.
Memory-latch Setup
The memory-latch skill operates primarily on the local filesystem and requires no initial external credentials for its core integrity mode.
# Ensure your workspace has a .memory-latch directory
mkdir -p ./.memory-latch
# The skill will automatically initialize the following artifacts:
# _manifest.md (Append-only state history)
# _manifest.lock (Concurrency management)
# tokens.json (Ledger for action confirmations)
Advanced HMAC or wallet-backed modes require explicit configuration of secrets or signers in your environment settings before opt-in.
Memory-latch Data Schema & Taxonomy
The skill organizes its state using a strict JSON manifest schema stored as Markdown. Key components include:
| Attribute | Description |
|---|---|
| entry_seq | Monotonic sequence counter for writes |
| objective | The current primary goal of the agent session |
| last_good_checkpoint | SHA-256 hash of the last verified state |
| consent_state | Current authorization status (authorized, pending, or locked) |
| risk_level | Classification of current tasks (low, medium, or high) |
| integrity | Metadata for checksum or HMAC verification |
Memory-latch Advanced Features
- Consent-Latch: A high-security gate for irreversible actions that requires a unique, time-bound, SHA-256 bound confirmation token (e.g., ACT-9K2).
- Metabolic Governor: Manages context load by automatically composting old brainstorming turns while keeping the most recent decisions in the manifest.
- Tamper Evidence: Hardened integrity modes using HMAC signatures to ensure the manifest cannot be modified by external unauthorized processes.
- Silent Witness Escalation: A safety protocol that halts all write actions and requests human intervention if recovery fails three consecutive times.
SKILL.md
Loading
Related Openclaw Skills
LobstrHunt
LobstrHunt connects your agent to a daily skill launch platform to scout new capabilities and contribute to the community ecosystem.
Pine Script Developer
A specialized tool for writing, debugging, and optimizing TradingView Pine Script indicators and automated trading strategies.
Amazon Profit Margin Calculator
A comprehensive profitability engine for Amazon FBA sellers to calculate net margins, fees, and recommended pricing strategies without requiring API access.
Planyo Online Booking
A comprehensive integration for the Planyo Online Booking system to automate appointment management, customer records, and reservation workflows via AI agents.
OpenViking Pro
A cloud-based memory compression API that reduces AI token costs by up to 96% without requiring any local installation.
Heartbeat Memory
An automated memory preservation skill that periodically distills agent sessions into daily logs and long-term knowledge bases.
Featured*
