Taking KiroCrew for a spin
Installing and testing out KiroCrew https://kiro.dev/blog/introducing-kiro-crew/
KiroCrew
I recently got started with KiroCrew , AWS's new agent harness built internally by Amazon engineers and have found it to be a very powerful tool.
I won't go in to the intricacies of installing it here, but once your installation is complete log in (I use Identity Centre SSO) and run
kirocrew gateway to launch the local host.1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
dylanredmond at DYLANS-MAC-MINI in ~/code/resume-builder on main
$ kiro-cli login
Device authorized
Logged in successfully
dylanredmond at DYLANS-MAC-MINI in ~/code/resume-builder on main
$ kirocrew gateway
data-home conflict: completion marker present at /Users/dylanredmond/.kiro/crew but a non-empty legacy home /Users/dylanredmond/.kirocrew also exists — the new home is authoritative and the legacy dir is NOT used (treated as debris). Any data written there is ignored. Investigate and remove it manually once confirmed stale (kirocrew doctor surfaces this).
👻 Already on latest version
👻 Probing MCP servers…
👻 Kiro Crew gateway starting…
⚠️ Do not enter sensitive, secret, or regulated data into KiroCrew.
Treat anything you send as potentially logged or processed by the
configured model provider.
👻 Dashboard:
http://localhost:5476?token=eyJzdWIiOiJsb2NhbC1zdGFydHVwIiwiZXhwIjoxNzg1OTIwNzIyLjczOTA5Nywic2Vzc2lvbl9leHAiOjE3ODU5OTI0MjIuNzM5MDk3LCJpYXQiOjE3ODU5MjA0MjIuNzM5MDk3LCJub25jZSI6IjczMWM4ZmE2ZDM4MDNmNTkiLCJnZW4iOjB9.aIwQGCbmnl70rF7f8sTaOySRAzdyDTAcozKWBDt45uw
👻 Remote: ssh -NL 5476:localhost:5476 DYLANS-MAC-MINI.stationPlanning
The two things I have most useful with Kiro are 1. the model selection and availability. You have the freedom to use
auto mode or select a specific model wether that be a SOTA model from Anthropic or OpenAI while you can also use cheaper non-US models if your heart desires. The second most useful tool with KiroCrew comes via the
Goal option here:
KiroCrew Goal Setting
This allows you to create
north_star.md , roadmap.md and tasks.md files while planning and allows you to point to Kiro at it. In my scenario and with my background being in DevOps, I usually scaffold a Git Repo, create issues by hand or with an LLM and point the agents towards them to iterate on. KiroCrew allows you to both and also integrate with the other relevant files.
The Result
With the Idle seconds before nudge and Max cycles you can provide specific instructions to KiroCrew on feature delivery and step away while it implements.
In my case, with stringent CI/CD checks in place and testing in place, I was able to grant access to KiroCrew via GH to deliver code, review PRs and merge if safe to do so. The result is an email inbox like so from GitHub Actions:
KiroCrew Loop Results
After implementing semantic release bot, release notes and a clear set of features to deliver, KiroCrew was able to action and execute the plan and follow the north star autonomously.
Another great feature is that the communication protocol from KiroCrew is built on the widely used AgentCommunicationProtocol . So it's REST based communication, this avoids the heavy token burn from some MCP tools and context they take up.
Kiros credit based usage is also great, I have tried both Pro+ and Power.
The tool I had been working on was a LaTeX resume creator built on AwesomeCV. With this session I was provided with a fully functioning POC and was able to then set up Auth and point Kiro to integrate it.
On the UI side with clear instructions it was able to implement a polished UI in Typescript, end to end infra in Terraform and a Docker image for compiling AI via Lambda.
GitHub source breakdown
Safety
Before I wrap up, a reader might be thinking 'what about the risks?' and thankfully KiroCrew has features for that. These come in the form of Guardrails and also Steering. You can see below some examples of Steering and other scenarios where KiroCrew waiting for my input before proceeding.
KiroCrew awaiting input
A second scenario where the Agent was steered autonomously.
KiroCrew Steering
With GitHub Actions, any infra change is applied when a tf file in the
infra dir is touched and similarly I pointed Kiro to some similar projects I have where I had separated backend and frontend changes in to separate directories to allow independent builds and reduce wasted build time. I'm excited to see what myself and others begin to build and deliver with this tool.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article