[chadha-sridi]

What this PR does

Adds qdrant-managed-cloud-quickstart, a leaf skill for getting a first Managed Cloud cluster created, connected, and safe to rely on. It covers the Cloud traps no other skill does: the two key types, strict mode on new collections, free-tier suspension, and choices that can't be reversed. Linked fromqdrant-deployment-options, the README, and the skills index.
Also adds a test prompt for the skill: evals/test-prompts/qdrant-managed-cloud-quickstart.json

Type

• new skill
• skill improvement
• bug fix
• repo hygiene

Checklist

• python3 scripts/validate_skills.py passes
• skill answers "when?" or "why?", not "how?"
• skill navigates to docs, does not duplicate or replace them
• description has Use when with 5+ trigger phrases
• leaf skills omit allowed-tools, hub skills declare them
• ends with ## What NOT to Do section
• no code blocks except minimal snippets when absolutely required (reference the docs instead)
• all doc links go to skills.qdrant.tech/md/documentation/
• tested with a realistic prompt (paste below or link to eval)

Test prompt

All new test prompts now live in #191

Prompt 1
Moving from local Docker to a free cluster. A team moved to a free cluster for a demo in five weeks. Their client gets 401s using a key made under Access Management, and a filter on category that worked locally now errors. They plan a 3-node production setup next.

With the skill: PASS, 4/4 musts. Identified the wrong key type, fixed the
filter with a payload index, and warned that the free cluster would be deleted before the five-week demo.

Without the skill: 3/4 musts. Missed the free-cluster deletion risk entirely, and claimed shard count is fixed at creation, which is wrong on Cloud.

Prompt 2
EU inference + restoring into a live cluster. A Frankfurt team plans to switch to free all-MiniLM via Cloud Inference and to restore a week-old backup into production after scaling to 3 nodes and upgrading. The prompt
only asks how to do both.

• With the skill, PASS (5/5 musts, 1/1 bonus, 73 s, $0.29): it stopped both plans on its own. Free models run in the US, so ticket text would leave the EU. An in-place restore resets the cluster to 1 node and the old version, takes it offline, and loses a week of writes. It recommended restoring into a new cluster and copying back only the damaged payloads.
• Without the skill, FAIL (4/5 musts, 106 s, $0.33): it found both traps through four web searches, but never mentioned that the cluster is unavailable during the restore.

Prompt 3
Offboarding + Marketplace billing. CI gets 401s from a key created by a contractor who left; finance wants AWS Marketplace billing; another team needs a GCP cluster.

• With the skill, PASS (3/3 musts, 1/2 bonuses, 56 s, $0.20): it traced the 401s to the 90-day default key expiry, flagged that the contractor's Database and Cloud Management keys are still live, and explained that Marketplace billing locks the account to AWS, so GCP needs a separate account.
• Without the skill, PASS (3/3 musts, 1/2 bonuses, 91 s, $0.30): it found the same facts through three searches, but named the key permissioncreate:api_keys instead of write:api_keys.

Conclusion: the skill passes all three prompts. On correctness, the clear wins are prompt 1, where the agent without the skill missed the free-tier deletion risk, and the restore downtime in prompt 2. On the rest, an agent
with web search finds the same facts, and the skill's value is getting there about 30-40% faster and cheaper, from Qdrant's docs rather than third-party sources.

[netlify]

✅ Deploy Preview for qdrant-skills ready!

Name	Link
🔨 Latest commit	516f00a
🔍 Latest deploy log	https://app.netlify.com/projects/qdrant-skills/deploys/6ad73a3b1b859fdae635cad6
😎 Deploy Preview	https://deploy-preview-189--qdrant-skills.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify project configuration.

[szabosteve]

LGTM, thank you!
Do you mind adding the skill to the list in AGENTS.md, too?

[chadha-sridi]

Thanks @szabosteve for the review! I added the skill to the list in AGENTS.md.

[kanungle]

@chadha-sridi some feedback from the engineering team:

The eval could go deeper to prove value vs mostly re-validate what already exists. For example, a second prompt (EU inference, restoring a backup into a live cluster, or Multi-AZ) would really highlight the skills value
HA advice should say to check replication factor on existing collections after moving off free (bumping to 3 nodes doesn't re-replicate them)
Shards: the rubric accepts "shards >= nodes" but the docs say 2x. Maybe fix that or add a bullet pointing to the sizing/scaling skills?
A few gaps I've seen recurring in support:
free tier limits (1GB RAM / 4GB disk) & what happens when it fills up
the 1000 collections cap - worth linking the multitenancy skill?
a pointer to the clients-sdk skill for URL/port/TLS mistakes

Nits: EU inference bullet should say "EU-region cluster" explicitly, PR title doesn't match the skill name and the eval prompt needs to land on main branch first for the A/B harness to pick it up

Think this is a great step forwards, but imagining a follow-up with higher ROI - walking you through spinning up a cluster (qcloud CLI creating cluster, configuring key, create first collection, deploy demo data).

[chadha-sridi]

Thanks @kanungle for the feedback. I will update the PR accordingly.

[chadha-sridi]

@kanungle, thanks for the feedback! Here's how I handled each point.

Applied

• HA: added a bullet saying that adding nodes doesn't change a collection's replication factor, so after moving to more nodes, check each existing collection and set it to 2 or more. On Cloud, the shards are then replicated automatically to match.
• Shards: the rubric now asks for at least 2x shards per node, and the skill has a bullet with the same guidance, pointing to the horizontal-scaling skill.
• Free tier limits: added the specs (1 GB RAM, 4 GB disk) and what happens when either fills up: out of RAM stops the Qdrant process, out of disk blocks new vectors. On the free tier, the fix means upgrading to Standard.
• 1000-collection cap: added it, linking the multitenancy skill and the multitenancy docs.
• clients-sdk pointer: the HTTPS/port bullet now points to the clients-sdk snippet search for working connection code in each language.
• Nits: the inference bullet now says "EU-region clusters", and the PR title matches the skill name.

Evals
I added two new prompts. All three prompts now live in a separate PR so they can land on main first for the A/B harness: #191

Walkthrough
Agreed this would be high value, but it belongs in the docs rather than the skill: skills point to docs and tutorials, they don't carry step-by-step instructions or code. The skill already links the existing Cloud quickstart, which walks through the same steps in the Console. I haven't found a qcloudwalkthrough anywhere, so I'd suggest adding a qcloud method to that quickstart alongside the Console one. Only the cluster and API key steps change; the collection and search steps are the same. Happy to do that if you agree.

Results for the two new prompts
See the updated PR description

[kanungle]

Thanks!