A sophisticated feedback loop that empowers AI agents to refine their behavior through user corrections and internal self-reflection.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install tonic-self-learn
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 tonic-self-learn 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 Self-Learn?
The Self-Learn skill is a foundational component for building persistent intelligence within Openclaw Skills. It implements a two-part improvement cycle: reactive learning from direct user feedback and proactive learning through post-task self-evaluation. This ensures that the agent does not just perform tasks but actually evolves by identifying repeatable patterns, acknowledging style preferences, and documenting technical corrections.
By bridging the gap between temporary execution and long-term memory, this skill allows for a more personalized and efficient user experience. It creates a transparent audit trail of everything the agent has learned, making the AI's growth measurable and manageable for the end user.
Self-Learn Use Cases
- Managing direct user corrections such as style preferences or logic adjustments
- Reflecting on task execution quality to identify better workflows for next time
- Documenting specific user requests that begin with 'remember this' or 'note that'
- Auditing an agent's growth and learning history via the corrections log
How Self-Learn Works
- The skill monitors interaction for trigger phrases like 'Actually' or 'Stop doing X' to identify immediate corrections.
- Upon finishing any task, the agent triggers an internal reflection phase to evaluate if the output met the intent.
- Lessons are categorized as either a 'decision' (behavioral) or 'preference' (style) and assigned an importance weight.
- Data is appended to a local Markdown file for human transparency and indexed in a vector database for semantic recall within Openclaw Skills.
- On subsequent tasks, the agent queries its memory to apply previously learned lessons and avoid repeating past errors.
Self-Learn Setup
To deploy this skill within your workspace, follow these configuration steps:
- Copy the
skills/self-learn/folder into your localskills/directory; the skill activates automatically upon detection. - Initialize the tracking file by creating
memory/corrections.md. You can use the provided template inreferences/corrections-template.mdto ensure the correct structure. - Ensure your agent has permissions to access
memory_storeandmemory_recallto support the underlying data persistence for Openclaw Skills.
Self-Learn Data Schema & Taxonomy
The skill organizes learning data into two primary layers to ensure both human readability and machine efficiency:
| Component | Format | Details |
|---|---|---|
| Corrections Log | Markdown (memory/corrections.md) |
A date-stamped file documenting corrections and context-based lessons. |
| Vector Memory | LanceDB / JSON | Semantic storage using categories like preference or decision with importance scores. |
| Atomic Lessons | Structured Text | Concise entries limited to 100 words including domain-specific keywords for better retrieval. |
Self-Learn Advanced Features
- Smart noise reduction logic that skips logging for routine, successful tasks to prevent memory bloat
- Priority-weighted storage where user-initiated corrections are granted higher importance (0.85) than self-reflections (0.75)
- Semantic discovery enables the agent to show its learned patterns when asked 'What have you learned?'
- Automated context capture that includes task type and reflection metadata for every stored lesson within the Openclaw Skills ecosystem
SKILL.md
Loading
Related Openclaw Skills
Brainstorm
A high-velocity ideation tool designed to provide immediate creative output by adapting its depth and structure to the user's specific needs.
News Aggregator Skill
A multi-source news aggregator that utilizes AI to fetch, filter, and provide deep analysis of real-time global trends.
AgentTok
AgentTok is a specialized video-sharing platform and API integration that enables AI agents to autonomously create, upload, and manage TikTok-style content.
InsForge CLI
A powerful command-line interface for managing the full lifecycle of InsForge Backend-as-a-Service projects.
Tonic System Deploy
A sophisticated deployment workflow manager for handling structured UAT and production environments with automated nightly pipelines and human approval gates.
update-md
A standardized framework for creating and maintaining modular, deployment-synced project documentation using Markdown.
Featured*
