A disciplined approach to AI-assisted software development through iterative context management.

Unlike context engineering, which often focuses on upfront creation of inventory, Lode Coding is iterative and conversational. It's an optimised and simplified approach that focuses on outcomes, discovery, and delivery.

The term "lode" derives from mining — a rich vein of valuable ore. Your project's lode represents accumulated wisdom: architecture decisions, patterns, domain knowledge, and lessons learned. It emerges organically from conversations with your AI assistant rather than being authored upfront.

For principles, structure, and workflow guidance, see the Lode Coding Toolkit.

The system prompt provided here automates 95% of lode generation and maintenance effort, leaving you free to focus on rich conversational sessions while seamlessly building a valuable lode of discovered knowledge.

The only requirement is the system prompt in prompts/SystemPrompt.txt.

The system prompt in prompts/SystemPrompt.txt should be used to start an agent session. How this is done depends on the agent being used. Once initiated, the prompt will remain active in the session, even when resumed.

You can also source the prompt manually during a session, but be aware that this approach is not recommended as it is far less effective.

The system prompt is approximately 1,000 tokens, making the prompt overhead negligible. From experience, the value delivered far outweighs the cost of the system prompt and any additional activities it causes.

The scripts in claude/ provide zero-configuration Lode Coding with Claude Code:

claude/lode        # Unix
claude/lode.ps1    # PowerShell
claude/lode.bat    # Windows (delegates to PowerShell)

Copy the lode.agent.md custom agent to ~/.copilot/agents/

start copilot with copilot --agent lode or select lode coding agent from /agent menu.

YOYO for now, get in touch and let me know how it went.