kiro-impl

Implements approved spec tasks using TDD with subagent dispatch and independent review.

Updated Jun 23, 2026
One-click install
npx skills add https://github.com/j5ik2o/marp-ai-base --skill kiro-impl-j5ik2o
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: kiro-impl
Source: https://github.com/j5ik2o/marp-ai-base/tree/main/.agents/skills/kiro-impl
Command: npx skills add https://github.com/j5ik2o/marp-ai-base --skill kiro-impl-j5ik2o

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Turning an approved spec into working code often breaks down due to skipped tests, unreviewed changes, and lost context across tasks. This Skill executes tasks from a Kiro-style spec (requirements.md, design.md, tasks.md) with enforced TDD, independent review, and structured failure recovery. ## Core Features & Use Cases - Autonomous sub-agent execution: Dispatches a fresh implementer sub-agent per task, followed by an independent reviewer that verifies the actual git diff against the spec before approval. - Enforced TDD and feature flags: Requires tests written before implementation (RED phase evidence) and a feature-flag protocol for behavioral changes. - Bounded failure recovery: Routes blocked or rejected tasks to a fresh-context debug sub-agent with root-cause classification, capped at two debug rounds before blocking the task. - Use Case: After approving a feature spec with $kiro-spec-tasks, run the implementation phase to have each sub-task implemented, reviewed, committed selectively, and marked complete in tasks.md automatically. ## Quick Start Ask the agent to implement all pending approved tasks for your feature, for example: run kiro-impl on the user-auth feature to execute its tasks with TDD and review.

Frequently Asked Questions about kiro-impl

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I implement approved spec tasks with TDD automatically?

Run the skill with a feature name and no task numbers to enter autonomous mode. It dispatches a fresh implementer sub-agent per sub-task, requires RED-phase test evidence, then sends the git diff to an independent reviewer before committing.

How do I run only specific tasks instead of all pending tasks?

Pass task numbers as the second argument, such as 1.1 or 1,2,3, to enter manual mode. Selected tasks are executed directly in the main context using the same RED, GREEN, REFACTOR cycle and review protocol.

What happens when a task implementation fails review or gets blocked?

Rejected tasks are re-dispatched to the implementer with feedback, up to two rounds. Persistent failures trigger a fresh-context debug sub-agent that classifies the root cause and returns a fix plan, with a maximum of two debug rounds before the task is marked blocked.

Can I resume an interrupted implementation session?

Yes, re-running the skill is safe because completed tasks are already checked off in tasks.md and committed to git. The controller re-reads tasks.md each iteration and picks up where it left off.

What are the prerequisites before running task implementation?

The spec files spec.json, requirements.md, design.md, and tasks.md must exist, and tasks must be approved in spec.json. If not, the skill stops and directs you to complete the requirements, design, and tasks phases first.