Guide · Agent skill
How to use the Agentic Engineering Workflow agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The agentic-engineering skill is a set of working rules, written in Japanese, that tells an AI coding agent how to run a software project. It is for developers who let agents write most of their code and want more control over review, cost and measurement. After this guide you will know what the skill changes in the agent's behaviour, how to install it, how to trigger it, and where its limits are.
Agentic Engineering Workflowaffaan-m/ECC · MIT Download the skill Quick start
- 1Download the ZIP. Get the skill ZIP from this page. It contains SKILL.md.
- 2Unzip into a skills folder. Place the folder in ~/.claude/skills/ for personal use or in .claude/skills/ inside a project.
- 3Or upload it in the Claude apps. Go to Settings → Capabilities → Skills and upload the skill there.
- 4Prepare your tests or evals. Make sure the project has tests or evaluations, because the evaluation-first loop depends on them.
- 5Describe the task naturally. Ask for a plan with small verifiable units and a clear definition of done.
- 6Check the log. Ask the agent to record model, tokens, retries, time and outcome per task, then review it.
What the agentic-engineering skill makes the agent do
An agent skill is a folder containing a SKILL.md file with instructions that Claude reads when the skill is relevant. This skill does not add tools or programs. It only changes how the agent works. The source frames the goal as using an AI agent for most implementation work while humans control quality and risk.
- Define what 'done' means before running anything.
- Split work into units that can each be verified on their own.
- Choose a model tier based on task complexity.
- Measure results with evaluations and regression checks.
Follow the evaluation-first loop
The skill asks the agent to define capability and regression evaluations first, then run a baseline and capture failure signatures before implementing. After the change, the evaluations are run again and the deltas are compared. The loop only works if your project already has tests or evaluations to run.
- Define capability and regression evaluations.
- Run the baseline and capture failure signatures.
- Implement the change.
- Re-run the evaluations and compare the deltas.
Split work into 15-minute units
The skill applies a 15-minute unit rule. Each unit should be independently verifiable, carry a single main risk, and have a clear completion condition. Smaller units make AI-generated changes easier to review and roll back.
Route tasks to cheaper or stronger models
The skill describes a routing policy by tier. Haiku is suggested for classification, boilerplate conversion and narrow edits. Sonnet is suggested for implementation and refactoring. Opus is suggested for architecture, root-cause analysis and multi-file invariants. The skill also says to escalate to a higher tier only when a lower tier fails on a clear reasoning gap. The skill describes this policy; it does not switch models for you.
Track cost and focus code review
For every task, the skill asks you to record the model, a token estimate, retry count, wall-clock time and success or failure. For review, it prioritises invariants and edge cases, error boundaries, security and authentication assumptions, and hidden coupling or rollout risk. If an automatic formatter or linter already enforces style, style-only differences should not take up review cycles.
Install the skill
Download the ZIP from this page and unzip it into ~/.claude/skills/ to make the skill available for all your projects, or into a project's .claude/skills/ folder to limit it to that project. You can also upload the skill in the Claude apps under Settings → Capabilities → Skills. The package contains one file: SKILL.md.
- Personal: ~/.claude/skills/
- Per project: .claude/skills/ inside the project folder
- Claude apps: Settings → Capabilities → Skills
Trigger the skill with natural requests
You do not need a special command. Describe the way you want the work done, and the agent can apply the skill when it is relevant. The skill's description mentions evaluation-first execution, decomposition and cost-aware model routing, so requests about those topics are the best match.
- Plan this feature as small units that can each be verified, and define done before coding.
- Before changing the code, write evals and run a baseline test.
- Which model tier should handle this: a boilerplate conversion or a root-cause analysis across several files?
- Review this AI-generated change, focusing on edge cases, error handling and security assumptions.
- Give me a per-task log of model, retries, time and outcome for this refactor.
What people use it for
- Planning a feature as a set of small units, each with a clear finish condition.
- Setting up a baseline test run before an AI agent changes existing code.
- Choosing a cheaper model for a narrow boilerplate edit and a stronger one for a multi-file root-cause analysis.
- Reviewing AI-generated pull requests with a focus on invariants, error boundaries and authentication assumptions.
- Tracking cost and retries across a refactor to see where the agent struggled.
Frequently asked questions
How do I install the agentic-engineering skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. You can also upload it in the Claude apps under Settings → Capabilities → Skills. The package contains SKILL.md.
What do I need to use this skill?
The source lists no specific requirements. It is a single SKILL.md file with no scripts or external tools. To get the benefit of the evaluation-first loop, your project needs tests or evaluations.
Is the agentic-engineering skill free?
The skill is published under the MIT licence, as listed on this page. Check the licence file in the source repository for the full terms.
Does the skill run scripts or use the network?
No. The skill runs no scripts and uses no network access. It only provides written guidance that changes how the agent behaves.
Will the skill pick the best AI model automatically?
No. It describes a routing policy by tier that the agent should follow, but it does not choose or switch models on its own.
Is there an alternative if I want a ready-made tool?
This skill is a workflow guide, not an app or program. If you need a ready-made tool, look for one that actually runs code or tests; this skill only shapes how the agent approaches the work.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits developers and teams who already let AI agents write much of their code and want a checklist for review, testing and cost tracking. It is written in Japanese and needs a real test setup to be useful. If you want a ready-made app or a project without tests, pick a different tool.
Download the skill