A cross-session analysis tool that synthesizes accumulated learning files into ranked promotion candidates for AI project instructions.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install learning-aggregator
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 learning-aggregator 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 Learning Aggregator?
Learning Aggregator serves as the critical inspect step in an AI agent's outer loop. While agents frequently log corrections and knowledge gaps during active sessions, these insights often remain as stagnant data. This skill bridges that gap by reading accumulated files in the .learnings/ directory, deduplicating patterns via unique keys, and computing recurrence across multiple sessions to identify high-value improvements.
By transforming a write-only log into a structured gap report, the skill enables developers to see exactly where their AI agents are struggling. It categorizes gaps into knowledge, tool, or skill deficiencies, providing a clear roadmap for updating project instruction files like CLAUDE.md or AGENTS.md. This ensures that a mistake made today becomes a permanent prevention rule for tomorrow, maximizing the efficiency of Openclaw Skills.
Learning Aggregator Use Cases
- Weekly review of accumulated AI learnings to maintain project instruction health.
- Pre-task auditing to check if a specific codebase area has recurring failure patterns.
- Post-sprint consolidation of findings from multiple AI-assisted development sessions.
- Verification of promotion-ready flags generated by self-improvement automated workflows.
- Deep-dive incident analysis using session traces to identify silent retry loops or user corrections.
How Learning Aggregator Works
- Scans the .learnings/ directory to parse LEARNINGS.md, ERRORS.md, and FEATURE_REQUESTS.md for metadata like Pattern-Keys and Recurrence-Counts.
- Groups entries by Pattern-Key to sum recurrences and identify distinct tasks where the pattern appeared.
- Applies promotion thresholds to classify patterns as Promotion-Ready (3+ recurrences across 2+ tasks) or Approaching.
- Categorizes the underlying issue into specific gap types such as Knowledge Gaps, Tool Gaps, or Reasoning Failures.
- Generates a comprehensive Gap Report with evidence, related files, and recommended actions for human or agent review.
Learning Aggregator Setup
To use the Learning Aggregator within your Openclaw Skills environment, ensure your agent has access to the .learnings/ folder in your workspace. You can invoke the analysis manually using the following commands:
# Run a standard analysis of local learning files
learning-aggregator
# Run a deep analysis including session traces from the last 7 days
learning-aggregator --deep --since 7d
# Filter analysis to a specific technical area
learning-aggregator --area backend
Learning Aggregator Data Schema & Taxonomy
The skill organizes data based on a strict metadata taxonomy parsed from markdown files. It produces a Gap Report that follows this structure:
| Field | Description |
|---|---|
| Pattern-Key | Unique identifier for deduplication (e.g., harden.input_validation) |
| Recurrence-Count | Total number of times the pattern was observed |
| Gap Type | Classification (Knowledge, Tool, Skill, Ambiguity, Reasoning) |
| Evidence | List of specific log IDs and summaries supporting the pattern |
| Recommended Action | Concrete steps to update instructions or tools |
| Tracker-ID | Persistent ID used to trace the pattern through the full improvement lifecycle |
Learning Aggregator Advanced Features
- Deep Trace Analysis: Integration with Entire to analyze raw session transcripts for silent failures like retry loops and user corrections.
- Multi-Source Merging: High-confidence pattern detection by overlapping explicit logs with implicit session trace data.
- Threshold Customization: Use the --min-recurrence flag to adjust the sensitivity of promotion logic for different project scales.
- Automated Handoffs: Outputs structured data compatible with harness-updater and eval-creator for end-to-end automation of Openclaw Skills improvement.
SKILL.md
Loading
Related Openclaw Skills
WeShop Baby Face Generator Skill
An AI-powered skill that predicts a baby's appearance by blending facial features from two parent photos via the WeShop OpenAPI.
Bikini Try-On OpenAPI Skill
An advanced AI-powered virtual try-on tool that allows users to realistically overlay bikini designs onto person photos using the WeShop OpenAPI.
Intent Framed Agent
A specialized agent protocol that formalizes coding intent and monitors for scope drift during implementation sessions.
Brat AI Image Generator Skill
An AI-powered image generation tool for creating custom Charli XCX-style brat album covers and memes via the WeShop OpenAPI.
Book Deep Reader
A professional-grade research agent that transforms books into structured teaching materials, critical evaluations, and actionable knowledge frameworks.
Lockheed Martin Knowledge Skill
A specialized Openclaw Skills module providing deep technical and business intelligence on Lockheed Martin, the world's leading defense and aerospace contractor.
Featured*
