HyperStack is an auditable agent provenance graph that provides a verifiable memory layer for AI agents to prove knowledge and coordinate without LLM overhead.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install hyperstack
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 hyperstack 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 HyperStack?
HyperStack serves as the foundational memory layer for the next generation of Openclaw Skills, enabling agents to build a verifiable provenance graph. It moves beyond simple text logs to a structured, typed graph where every piece of information is timestamped, auditable, and linked with deterministic trust. This ensures that agents can prove what they knew and why they knew it at any specific point in time.
By utilizing HyperStack within your Openclaw Skills workflow, you ensure that agent decisions are traceable, conflicts are detected automatically, and coordination happens through typed signals rather than expensive LLM-in-the-loop processing. The system provides three distinct memory surfaces: episodic for session history with soft decay, semantic for permanent facts, and working memory for short-term scratchpad tasks.
HyperStack Use Cases
- Tracking complex decision rationale and detecting hindsight bias in retrospective audits.
- Coordinating multi-agent systems using typed signals and shared workspace identity.
- Mapping dependencies and impact blast-radius before agents execute high-risk code changes.
- Restoring session context via episodic memory traces with automatic time-based decay.
- Implementing verifiable RAG where every retrieved fact includes confidence scores and truth stratum.
How HyperStack Works
- The agent identifies itself via the identity tool to establish a SHA256 fingerprint and trust score for all subsequent actions.
- Information is ingested into the graph as typed cards with specific relations like depends-on, triggers, or blocks.
- Smart search automatically routes queries to the most efficient retrieval mode, such as semantic search, impact analysis, or dependency lookups.
- Conflict detection and staleness cascades ensure the graph remains logically consistent as data evolves, flagging contradictions automatically.
- Decision replay allows for time-travel to past graph states to audit exactly what an agent knew at any specific moment, identifying potential hindsight bias.
HyperStack Setup
To integrate HyperStack into your Openclaw Skills environment, add the MCP server configuration to your client.
# Install the core library
npm install hyperstack-core
# Or for Python users
pip install hyperstack-py
Update your mcpServers config:
{
"mcpServers": {
"hyperstack": {
"command": "npx",
"args": ["[email protected]"],
"env": {
"HYPERSTACK_API_KEY": "your_api_key",
"HYPERSTACK_WORKSPACE": "project_name"
}
}
}
}
HyperStack Data Schema & Taxonomy
HyperStack organizes data into cards within a provenance graph with the following metadata taxonomy:
| Field | Description | Type |
|---|---|---|
| slug | Unique identifier for the card | String |
| title | Human-readable name of the card | String |
| cardType | Category like decision, person, workflow, or project | Enum |
| memoryType | Surface: working, semantic, or episodic | Enum |
| confidence | Self-reported certainty (0.0 to 1.0) | Float |
| truthStratum | Epistemic status: draft, hypothesis, or confirmed | String |
| links | Typed relations to other cards (e.g., slug:relation) | String |
HyperStack Advanced Features
- Git-style memory branching allowing for safe experimentation through forks, diffs, and merges.
- Deterministic agentic routing that selects the optimal retrieval mode without incurring LLM costs.
- Decision replay with hindsight bias detection to flag cards modified after a decision was made.
- Utility-weighted edges that automatically promote high-value information based on agent feedback loops.
- Staleness cascades that propagate through the graph to flag dependent nodes when parent information changes.
SKILL.md
Loading
Related Openclaw Skills
Void Atlas Protocol
A four-axis ethical navigation framework for evaluating systems, agents, and policies using quantitative waypoints and routing rules.
OpenClaw Flow Kit
A comprehensive utility suite designed to resolve workflow bottlenecks, handle rate limits, and standardize script execution for AI agents.
LYGO-MINT Operator Suite
A technical toolkit for canonicalizing, hashing, and anchoring multi-file AI agent packs into verifiable, receipts-first ledgers.
Book Brain Visual Reader
A cognitive architecture for AI agents that synchronizes visual perception with a structured 3-brain memory system.
SkillFence
A dedicated runtime security watchdog designed to monitor the activity of Openclaw Skills by tracking network connections, process behavior, and credential access.
ZeroRules
An AI agent skill that intercepts deterministic tasks like math and time queries locally to eliminate unnecessary LLM token costs.
METADATA
Featured*
