An advanced context management system that uses structured snapshots to ensure AI agents never lose track of complex coding tasks during context compaction.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install big-memory
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 big-memory 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 Big Memory?
Big Memory is a specialized framework designed to solve the problem of context loss in long-running AI development sessions. By implementing a three-phase protocol—Capture, Detect, and Recover—it allows Openclaw Skills to maintain a high level of continuity even when the underlying LLM reaches its token limit and triggers a memory flush. It creates machine-parseable delimiters within your project memory that act as a durable anchor for active goals, code context, and next steps.
This system ensures that the agent remains productive without requiring the user to manually re-explain the current state of work. The protocol is entirely self-contained, utilizing built-in tools like memory search and file editing without needing external databases or API keys, making it a robust addition to any professional development workflow using Openclaw Skills. It effectively transforms volatile short-term context into structured long-term knowledge.
Big Memory Use Cases
- Protecting complex debugging state before context compaction occurs during long sessions.
- Resuming work seamlessly after the agent forgets specific file paths or variable names.
- Creating a reliable checkpoint before pivoting to a different subtask or branch.
- Manually saving task progress using the /big-memory save command for multi-day projects.
- Coordinating between different agents by providing a structured hand-off snapshot in the shared memory.
How Big Memory Works
- The agent monitors for system compaction signals, user commands, or significant work milestones to trigger a capture phase.
- A structured task snapshot is generated following a strict schema including active goals, files in play, decisions made, and specific code context.
- This snapshot is appended to the daily memory log (memory/YYYY-MM-DD.md) using unique HTML comment markers that enable precise machine retrieval.
- Upon detecting context loss or receiving a /big-memory recall command, the agent initiates a search for the BIG-MEMORY-SNAPSHOT marker.
- The agent parses the most recent snapshot, presents the recovered state to the user for validation, and resumes work from the exact next steps defined in the snapshot.
Big Memory Setup
To integrate Big Memory with your environment, ensure your project includes a memory directory. You can optimize automatic snapshots by updating your openclaw.json configuration to use the Big Memory system prompt for memory flushes.
{
"agents": {
"defaults": {
"compaction": {
"memoryFlush": {
"enabled": true,
"softThresholdTokens": 40000,
"systemPrompt": "You are capturing structured task state for post-compaction recovery. Follow the BIG-MEMORY-SNAPSHOT schema exactly."
}
}
}
}
}
Use the following command to manually save your first checkpoint within your Openclaw Skills workflow:
/big-memory save
Big Memory Data Schema & Taxonomy
The skill organizes task data into a standardized Markdown schema wrapped in HTML comments for high-relevance search matching:
| Section | Description |
|---|---|
| Metadata | Contains timestamp, snapshot-id, and version markers for machine parsing. |
| Active Goal | A concise one-sentence description of the user's primary objective and the 'why' behind it. |
| Current State | Tracks the current phase (planning, debugging, etc.), git branch, and progress percentage. |
| Files In Play | A list of specific file paths and their roles in the current active task. |
| Decisions Made | A log of technical decisions and their rationales to prevent repetitive questioning. |
| Code Context | Essential signatures or snippets that cannot be easily reconstructed from files. |
| Next Steps | An ordered, actionable list of the immediate tasks required to complete the goal. |
Big Memory Advanced Features
- Automated pre-compaction flushes that replace generic summaries with actionable, structured recovery data.
- Machine-parseable HTML delimiters that optimize BM25 search relevance for Openclaw Skills memory tools.
- Milestone-based self-assessment where the agent proactively snapshots context-heavy logic before it is lost.
- Multi-snapshot versioning that allows agents to roll back to earlier task states if the latest context becomes stale.
- Built-in status commands to monitor snapshot density and divergence from the current task state.
SKILL.md
Loading
Related Openclaw Skills
Sendook Email Skill
A specialized skill for AI agents to interact with Sendook inboxes by reading, searching, replying to, and sending emails.
Sendook Email Skill for Openclaw
A specialized skill for AI agents to interact with existing Sendook email inboxes for automated messaging and thread management.
GTA Property Investment Report Generator
An automated tool for generating professional property investment financial reports specifically for the Greater Toronto Area.
Switchboard Data Operator
An autonomous operator for designing, simulating, and deploying Switchboard on-demand data feeds and randomness across multiple blockchain ecosystems.
Numerai Tournament
Automate participation in the Numerai hedge fund tournament to train models and earn NMR cryptocurrency.
ObekT Security
A comprehensive threat detection and security analysis toolkit designed to audit code, files, and AI agent skills for vulnerabilities and malicious patterns.
Featured*
