Your team already uses AI to write code. DevTeam AI Kit handles the rest of the job: reviews, specs, tickets, tests, PR descriptions, postmortems, release notes, 1:1s, sprint planning and hiring.
One-time payment · Installs in 2 minutes · 14-day refund
Each skill is a tested workflow in plain Markdown. It reads your repository, your conventions and your templates before it writes anything.
code-reviewSenior-level review. Every finding is verified, located and ranked by severity.
ticket-to-specTurns a vague request into a spec grounded in your real code, with honest open questions.
task-breakdownSplits a spec into small, ordered, shippable tickets for GitHub, Jira or Linear.
test-writerWrites meaningful tests in your project's own style, and runs them until green.
pr-descriptionPR title and body from the actual diff, using your PR template.
adr-writerArchitecture Decision Records with options, trade-offs and sources.
incident-postmortemBlameless postmortem with timeline, root causes and owned action items.
release-notesA developer changelog plus a plain-language announcement for users.
one-on-one-prep1:1 prep sheet from recent work and your previous notes. No productivity scores.
sprint-planningCapacity, backlog readiness check and a realistic sprint proposal.
interview-kitStructured interviews: competency grid, questions, exercise and rubric.
We added a 9-line endpoint with four hidden problems to a test repository and typed review my changes. This is a shortened excerpt of what the code-review skill returned.
The change
// src/orders.js
async function listOrders(req, res) {
const status = req.query.status;
const orders = await db.query(
`SELECT * FROM orders
WHERE status = '${status}'`);
for (const o of orders) {
o.items = await db.query(
'SELECT * FROM items WHERE order_id = $1',
[o.id]);
}
const total = orders.reduce(
(s, o) => s + o.amount, 0.0);
res.json({ orders, total });
}
The review
Verdict: 🛑 Request changes 🛑 Blocking 1. SQL injection - src/orders.js:12 Scenario: GET /orders?status=' OR '1'='1 returns every order. Fix: use a parameterized query, as getOrder does. 2. Missing authorization: returns other users' orders - src/orders.js:12 The query has no user_id filter. getOrder checks it, listOrders skips that check. ⚠️ Should fix 3. N+1 queries - src/orders.js:13-15 5,000 matching orders means 5,001 sequential round trips. 4. Money summed as floats - src/orders.js:16 Sum in SQL or use integer cents. 👍 Good The items query on line 14 is parameterized correctly.
You can. This kit saves you the weeks of iteration it takes to make them reliable for a whole team.
Skills read your repo, linters, PR templates and existing tests first. They follow your style instead of imposing one.
One team-kit.md file holds your stack, commands, conventions and rituals. Commit it and everyone gets the same behaviour.
Skills separate facts from assumptions, try to disprove findings before reporting them, and ask instead of inventing.
Nothing is posted, merged, tagged or created in your ticket tool without your confirmation.
Skills are plain Markdown files in the standard SKILL.md format. No dependency, no server, no account.
Copy the folder
cp -r devteam-ai-kit/skills/* \ .claude/skills/
Describe your team
cp team-kit.example.md \ .claude/team-kit.md
Ask Claude Code
> review my changes > write the PR description
One-time payment. Free updates for version 1.x. 14-day refund if the kit is not for you.
For one developer or tech lead
For one team or company, up to 10 seats
Launch prices. Secure checkout and VAT handled by our payment partner.
Claude Code (CLI, desktop app or IDE extension) on any plan that includes it. Most skills also use git and the GitHub CLI gh. Jira and Linear features need the matching connector.
The kit is built and tested for Claude Code. The files are plain Markdown, so you can use them as rules or prompts in other tools, but results will vary.
The skills are stack-agnostic. They detect your framework and test runner from the repository and follow the conventions they find there.
No. The kit contains only instructions for your own Claude Code session. It has no server and no telemetry.
Yes, for your own or your team's use. Reselling or publishing them is not allowed.
Full refund within 14 days of purchase, no questions asked. See the refund policy.
Email support. We answer within two business days.