A specialized spaced repetition workflow that enables AI agents to maintain high-fidelity long-term memory through structured daily reviews and self-grading.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install srsa
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 srsa 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 SRSA Review Skill?
The SRSA Review Skill implements Spaced Repetition Systems for AI Agents (SRSA) to bridge the gap between ephemeral context and long-term knowledge retention. By utilizing a command-line driven interface, this skill allows agents to manage specialized cards that represent user preferences, corrected actions, and critical decision-making logic. It is a vital component of the Openclaw Skills ecosystem for developers building persistent, self-improving AI entities.
This skill enforces a rigorous review loop where the agent must recall information from its internal memory before verifying against a reference answer. It transforms the passive storage of data into an active maintenance process, ensuring that the agent's internal memory remains accurate, relevant, and highly retrievable over time.
SRSA Review Skill Use Cases
- Reinforcing user-specific preferences and style guidelines over long periods.
- Correcting agent behaviors that have been previously flagged or edited by the user.
- Managing complex decision-making logic that the agent initially hesitated to execute.
- Implementing a daily maintenance routine to keep the agent's knowledge base optimized.
- Explicitly updating or deleting interfering memories that cause hallucination or confusion.
How SRSA Review Skill Works
- The agent initiates the session by checking the review status to determine the number of due cards.
- The agent fetches a question using the get-question command and attempts to provide an answer based solely on its current internal state.
- The reference answer is retrieved, and the agent performs a comparative analysis between its recall and the truth.
- A self-rating (again, hard, good, or easy) is submitted, which updates the SRSA scheduling algorithm for that specific card.
- If the recall was not perfect, the agent follows a mandatory reflection template to identify specific memory updates (add, delete, or update).
- The loop continues until the system reports no due cards remain for the day.
SRSA Review Skill Setup
To integrate this into your agent environment, ensure you have the SRSA scripts available. Use the following commands to get started:
# Check the current status of the memory deck
uv run python scripts/main.py status
# Create your first memory card
uv run python scripts/main.py card new -q "User Preference" -a "The user prefers Python for all scripting tasks."
# Start the review process
uv run python scripts/main.py review get-question
SRSA Review Skill Data Schema & Taxonomy
The skill manages information through a structured lifecycle of cards and review states. Data is organized as follows:
| Component | Description |
|---|---|
| SRSA Cards | Managed objects containing questions, answers, and scheduling metadata. |
| Rating State | Sequential enforcement (Question -> Answer -> Rate) to prevent state errors. |
| Reflection Metadata | Agent-generated conclusions used to trigger explicit updates to the primary memory system. |
| Retrievability Stats | Metrics tracking average accuracy and future due dates. |
SRSA Review Skill Advanced Features
- Manual card overrides to fix ambiguous prompts or outdated reference information during the review loop.
- Explicit memory system hooks where the agent must state 'Add/Delete/Update' actions for external synchronization.
- Recovery protocols for handling command errors or invalid card IDs without breaking the review session.
- High-density status reporting including historical accuracy and future workload forecasting.
- Context-aware reflection limits to prevent long-running review sessions from polluting the agent's active context window.
SKILL.md
Loading
Related Openclaw Skills
Reach Agent Web Interface
Reach is a comprehensive web interface layer that allows AI agents to interact with the internet as naturally as humans do.
Manim Animation
A powerful skill for generating mathematical animations with integrated text-to-speech narration and synchronized subtitles.
Arabic AI Video Editor
An AI-powered video editing interface that allows users to edit and produce videos using natural language commands in Arabic and English.
Netlify
A comprehensive integration for managing Netlify resources, deployments, and DNS records through a secure, automated workflow.
AgentScope AI Agent Skill
AgentScope is a production-ready, open-source framework designed to build, evaluate, and deploy sophisticated multi-agent applications using large language models.
Traffic Quality Audit
A precision diagnostic tool for identifying low-intent traffic, bot spikes, and inefficient ad placements to reclaim wasted marketing budget.
Featured*
