A Claude Code skill, Claudeception, addresses the inefficiency of AI coding agents starting each session from scratch. It enables Claude Code to continuously learn by automatically saving discovered knowledge, debugging techniques, and workarounds as reusable "skills." This enhances agent intelligence and reduces repeated problem-solving time for users.

How It Works

The system leverages Claude Code's native skills framework. When Claude Code identifies non-obvious solutions or project-specific patterns, Claudeception captures this knowledge. It then formats it into a new skill—a markdown file with structured YAML frontmatter—optimized for future retrieval. During subsequent sessions, Claude Code loads these skills, allowing it to access and apply past discoveries contextually, rather than re-learning. This approach is inspired by academic research in AI skill libraries and self-reflection mechanisms.

Quick Start & Requirements

Installation involves cloning the repository to a local skills directory (e.g., ~/.claude/skills/claudeception) and setting up an activation hook script (~/.claude/hooks/claudeception-activator.sh). This script must be added to Claude's settings (~/.claude/settings.json) under UserPromptSubmit hooks to ensure continuous evaluation for extractable knowledge. The primary dependency is a functional Claude Code environment.

Highlighted Details

• Automatic Learning: Skills are generated automatically upon discovering non-obvious solutions, workarounds, or project-specific patterns during debugging or task completion.
• Explicit Control: Users can trigger a learning retrospective or explicitly save learned knowledge using the /claudeception command.
• Quality Gates: Extraction is selective, prioritizing knowledge that required genuine discovery, offers future utility, has clear triggers, and has been verified.
• Skill Format: Skills are stored as markdown files with YAML frontmatter, detailing the problem, context, solution, and verification steps.
• Research-Backed: Incorporates concepts from academic work like Voyager, CASCADE, SEAgent, and Reflexion on AI skill acquisition and self-improvement.

Maintenance & Community

The project is maintained by blader. No specific community channels (like Discord or Slack), sponsorships, or detailed roadmap information are provided in the README.

Licensing & Compatibility

The project is released under the MIT license. This permissive license allows for broad compatibility, including commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

Claudeception is dependent on the Claude Code environment and its native skills system. Its effectiveness relies on Claude Code's ability to identify and articulate discoverable knowledge. The README does not provide performance benchmarks or details on potential edge cases for skill retrieval or generation.