The foundational template for Self-Evolving Agent Skills (SEAS).
Standardized for autonomous learning, metacognition, and safety.

SEAS Core enables AI agents to transcend static instructions by implementing a Metacognition Loop:

• Journal: Record objective facts and subjective reasoning.
• Meditate: Analyze history to identify growth patterns.
• Evolve: Self-patch instructions and scripts.
• Verify: Guarantee system integrity via automated testing.

The following structure defines a SEAS-compliant skill. Items marked as Mandatory are required for the self-evolution engine to function.

seas-core/
├── 📄 SKILL.md            [Mandatory]  # Primary instructions & Evolution metadata
├── 📁 scripts/            [Mandatory]  # Logic for meditation, verification, and archival
├── 📁 journals/           [Mandatory]  # Short-term memory (Cases & Experiences)
├── 📁 archives/           [Mandatory]  # Long-term memory (History & Snapshots)
├── 📁 references/         [Mandatory]  # Knowledge base and specifications
│   └── 📘 Meta-Specification.md         # The blueprint for self-evolution
└── 📁 assets/             [Optional]   # Static resources (images, icons, etc.)

1. Real-time Journaling: Every failure or complex success is logged in journals/.
2. Periodic Meditation: When the threshold is met, the agent aggregates and analyzes journals.
3. Verified Evolution: The agent applies improvements and verifies them against scripts/verify_integrity.py.
4. Permanent Archival: Processed journals are archived into archives/ with a full rollback snapshot.

For a deep dive into the architectural constraints, safety protocols, and implementation details, please refer to the:

👉 Meta-Specification (Internal Link)

This project is a living experiment, co-architected by a human developer and the Gemini CLI Agent. It demonstrates the power of agent-led system evolution.

Licensed under the Apache License, Version 2.0. See the LICENSE file for details.