The only blueprint for long-running agent coding

Blueprint for Long-Running Agentic Coding Tasks

Eliciting Intent

• The first step in the coding process is to have the agent elicit your intent regarding the project and its goals.

• Utilizing tools like the slash grill me skill can help ensure that the agent fully understands your intentions through a series of probing questions.

Research and Master Plan Synthesis

• Once intent is established, the agent should conduct thorough research by pulling external documentation relevant to every aspect of the project.

• This phase culminates in synthesizing a master plan that outlines steps from A to Z, though it doesn't need to be overly detailed.

Skills Development and Spec Document Creation

• With intentions and research documented, these can be transformed into skills for dynamic information retrieval during project execution.

• The master plan should then be divided into smaller spec documents, each manageable within one agent's context window before resetting.

Inter-Spec Contracts and Validation Loops

• Each spec document acts as a puzzle piece with contracts ensuring dependencies are met; one spec must complete its tasks before another begins.

• A validation and testing loop is essential within each spec document to confirm that completed features do not introduce errors or regressions.

Framework Overview and Resources

• This framework serves as a generalizable approach for long-running coding tasks; it can be formalized into reusable skills.

• Access to additional resources, including the M2C1 autonomous agentic coding framework skill and related courses, is available via links provided in bio.