issue-driven-harness

Issue-driven development from a GitHub issue or Notion task starts by providing the issue URL, backlog key, or local Markdown file to launch a harness flow that coordinates Planner, Reviewer, Generator, and Evaluator sub-agents for iterative design and implementation.

A harness-based sub-agent workflow orchestrates specialized agents like Planner, Reviewer, Generator-FE, Generator-BE, and Evaluator to produce design artifacts and verify completion criteria through iterative planning and implementation loops, ensuring high-precision development on complex requirements.

Yes, you can use a local Markdown specification file to initiate the planning and implementation loop. The harness flow accepts local Markdown files alongside GitHub issues, Backlog tickets, and Notion pages as valid starting points.

Managing large-scale software design and verification across multiple agents is best handled by a harness flow that coordinates Planner, Reviewer, Generator-FE, Generator-BE, and Evaluator sub-agents to run iterative planning and implementation loops for quality assurance.

Yes, you need harness prompts located at autonomous-ai-agent-development/issue-driven/master-prompt-harness.md and canonical templates in ~/.agent/templates to drive the planning and verification loops of the issue-driven development workflow.

You should not use a sub-agent harness approach for small-scale tasks or when low token consumption is a priority, as coordinating multiple sub-agents like Planner and Reviewer involves high token consumption; use a simpler issue-driven start approach instead.