How to Submit a Skill to Claude
A step-by-step guide for Nexthink employees — from writing your first skill to seeing it live on Claude.
Last updated: June 2026 · Platform: Claude (Anthropic) · Owner: IT Infrastructure Team
1. What is a "skill" in AI?
In AI terminology, a skill is a packaged, reusable set of instructions that extends what an AI model can do. Think of it as a recipe you hand to Claude: it tells the model how to behave, what persona to adopt, which steps to follow, and which constraints to respect — every time that recipe is activated.
Everyday analogy
A skill is to Claude what a macro is to Excel. Once written and approved, anyone can invoke it to get consistent, governed behaviour — without retyping instructions every time.
Skills are plain-text or Markdown files (.txt or .md). A typical skill contains:
- A persona or role — e.g. "You are a senior Finance Analyst…"
- A task description — what the model should do when invoked
- Constraints / guardrails — what it must not do
- Output format instructions — structured JSON, bullet list, formal report…
- Examples (optional) — few-shot examples that improve consistency
Why every skill goes through a review
A skill runs inside Claude under your corporate identity. A malicious or poorly written skill could leak data, call internal APIs unintentionally, or bypass safety controls. The review process protects both you and the company.
2. Before you start
Make sure you have:
- Access to the AI Skill Validator app in Okta. If you don't see it in your Okta dashboard, request access through ServiceNow.
- Your skill written as a
.txtor.mdfile (max 10 MB), or ready to paste as plain text. - A clear, accurate description of what the skill does.
- A version number — start with
1.0.0for a new skill.
Writing tips for a faster review
Be explicit. Avoid vague directives like "be helpful". Instead write: "Respond only in English, keep answers under 200 words, and never disclose internal system names." Clear, unambiguous instructions pass the AI safety check more reliably.
3. How to submit a skill
Step 1: Open the submission form
Click Skills in the left navigation, then + Submit Skill in the top-right corner.
Step 2: Select the Claude platform
On Step 1, click the 🟠 Claude — Anthropic card, then Continue →.
Step 3: Fill in the skill details
| Field | Required | Notes |
|---|---|---|
| Name | Yes | Short descriptive title. |
| Description | No | One-paragraph summary of what the skill does. |
| Version | Yes | Defaults to 1.0.0. Must be changed on every update. |
| Tags | No | Comma-separated keywords to help others find the skill. |
| Skill content | Yes | Paste text directly or upload a .txt / .md file (max 10 MB). |
Step 4: Submit
Click Submit. You are redirected to the skill's detail page. The automated AI review starts within seconds — you can watch the status update in real time.
4. The review process
Every skill passes through an automated safety pipeline before it reaches Claude. The path taken depends on the AI's decision: whether it rejects the skill outright, escalates it for human review, or approves it directly.
Stage 1 — Automated AI Review
As soon as you submit, an automated workflow analyzes your skill for:
- Clarity, completeness, and professional quality
- Security threats: prompt injection, data exfiltration, jailbreak attempts, hidden payloads, scope creep
- Whether the skill references a Crown Jewel service (looked up live in ServiceNow)
The workflow produces an AI score (0–100) and a plain-language AI summary, both visible on the skill's detail page.
Stage 2 — Security Review (when escalated)
What is a Crown Jewel?
A Crown Jewel is a business-critical internal system that Nexthink has classified as high-risk. These are applications or services where unauthorised access, data leakage, or misuse could have a significant impact on the company — for example identity and access management platforms, financial systems, HR data, or core infrastructure. If your skill mentions or interacts with one of these systems, the AI detects it automatically and escalates to the Security team for review.
A skill is escalated to the Security team if either of the following is true:
- A Crown Jewel service is detected in the skill content, or
- The AI's confidence score is 90 or below out of 100 (indicating ambiguity or uncertainty)
The Security team receives a notification in Microsoft Teams and reviews the skill manually in the portal, then either:
- Approve — a ServiceNow ticket is opened automatically and IT is notified.
- Reject — the skill is marked with reviewer notes explaining why.
Only skills with no Crown Jewel and an AI confidence score above 90 skip this stage — a ServiceNow ticket is opened automatically without any human review.
Stage 3 — IT deploys to Claude
Once a ServiceNow ticket exists (from either approval path), the IT team receives the ticket and manually deploys the skill to Claude. The skill's status moves to Published.
Why a manual step? This manual step exists because Claude does not currently provide an API or automation to upload skills programmatically, so IT deploys each approved skill by hand.
5. Tracking your submission
From Skills in the portal, filter the list by any status at any time. Admins and Security roles see all skills; regular employees see only their own submissions.
Each skill's detail page shows:
- The current status badge (updates in real time)
- The AI score and AI summary once review completes
- The Crown Jewel flag and list of affected services if detected
- The ServiceNow ticket number with a direct link
- A full event timeline: submitted → AI review → (security review) → approved → published
6. Managing a published skill
| Action | Who | What happens |
|---|---|---|
| Request Update | Owner or Admin | You must bump the version number. A ServiceNow ticket is opened; an Admin confirms before the skill re-enters the review pipeline. |
| Request Deletion | Owner or Admin | A ServiceNow ticket is opened. An Admin must confirm before the skill is permanently removed. |
| Cancel Deletion | Owner or Admin | Cancels an in-progress deletion request and restores the skill to Published. |
| Withdraw | Owner | Immediately withdraws the skill while it is still under review, before publication. |
7. Status reference
| Status | Meaning |
|---|---|
| Pending | Submitted and queued — AI review has not yet started. |
| AI Review | The automated workflow is currently analyzing your skill. |
| AI Rejected | Failed the AI check. Read the AI summary on the detail page for the reason. |
| Security Review | Crown Jewel detected — the Security team is reviewing your skill. |
| Security Rejected | Rejected by the Security team. Check the reviewer notes on the detail page. |
| AI Approved | Passed the AI check. ServiceNow ticket opened — waiting for IT to deploy. |
| Security Approved | Approved by Security. ServiceNow ticket opened — waiting for IT to deploy. |
| Published | Live and available on Claude. 🎉 |
| Withdrawn | You pulled the skill back before it was published. |
| Update Requested | A new version is staged and waiting for Admin + IT confirmation. |
| Deletion Requested | Removal is pending Admin confirmation. |
8. Frequently asked questions
Q: My skill was AI-rejected. What should I do?
Open the skill's detail page and read the AI summary. It explains the specific reason (quality issue, security concern, ambiguous intent, etc.). Edit your content to address the feedback, bump the version number, and resubmit.
Q: How long does the whole process take?
The automated AI review completes in approximately 1 minute. If escalated to Security Review, timing depends on team availability. IT deployment follows the standard ServiceNow SLA once the ticket is assigned.
Q: Can I edit my skill while it is under review?
No. Editing is blocked while a skill is in AI Review or Security Review. Use Withdraw to pull it back, make your changes, and resubmit.
Q: Who can see my skill before it is published?
Only you, Admin roles, and Security roles can see your skill while it is under review. Once Published, it is available on Claude to all authorised users.
Q: What triggers a Security Review?
Two conditions can escalate a skill to the Security team: (1) a Crown Jewel is detected — a business-critical internal service (identity platforms, financial systems, core infrastructure) catalogued as high-risk in ServiceNow; or (2) the AI's confidence score is 90 or below out of 100, meaning the submission was ambiguous or unclear. Both are evaluated automatically by the AI workflow.
Q: How do I update an already-published skill?
Open the skill's detail page, click Request Update, increment the version number, update the content, and submit. A ServiceNow ticket is opened; once an Admin confirms, the skill re-enters the review pipeline from the beginning.
For support, raise a ticket in ServiceNow for the IT Infrastructure Team.