kiro-impl

You can automate test-driven development by running a skill that processes numbered tasks from tasks.md, enforcing RED to GREEN TDD cycles with automated implementer and reviewer subagents. It selectively stages changed files and generates structured status reports.

The RED to GREEN cycle requires writing failing tests first, then implementing code to pass them. Subagents enforce this by dispatching separate implementer and reviewer agents per task, parsing structured VERDICT blocks to gate progress before staging commits.

Yes, you need tasks.md, requirements.md, and design.md files defining numbered tasks, along with a spec.json for approval validation. The workflow applies to feature implementation scenarios where these canonical files define tasks and reviewers are required.

Yes, you can run all pending tasks as one-task-per-subagent autonomous iterations, or execute selected tasks manually in the main context. Both modes enforce TDD cycles, selective git staging, and structured review reporting.

Preventing regressions requires enforcing test-driven development cycles, using feature-flag patterns for behavioral changes, and gating commits behind reviewer subagent approvals. The workflow discovers canonical test and build commands for preflight validation before staging.

A debugger subagent with bounded remediation rounds prevents infinite loops when tests fail unexpectedly. It parses structured STATUS blocks from the implementer and applies bounded debug rounds to resolve issues before the reviewer subagent evaluates the final VERDICT.