Skills graph

This starter graph connects 19 skills. It does not replace the 26 learning paths. Mark a skill only after you can do the outcome, not after you open the page.

Confidentiality, integrity, and availability: Explain which security goal a simple control protects, without memorising a slogan. How a name becomes an address: Describe a normal DNS lookup well enough to notice a broken or unexpected answer. Read an HTTP status before guessing: Use the method and status code to say what the server claims happened. Object-level authorization: Show why a valid login does not grant access to every record, and name the trust boundary. Review a trust boundary before patching: Separate caller-controlled fields from server-resolved identity, then describe a fix and a residual risk. Triage a log before escalating: Say what a log line supports, what it does not, and which question to ask next. Keep an agent inside its tool boundary: Reject a model-supplied tool call that changes the tenant, object, or action. See a query change because of input: Show a request that changes a database query, and name the fix as parameter binding rather than a longer password. Read a token before you trust a claim: Point at a claim inside a token and say why the signature, not the claim text, is what a verifier must check. Name the signal in a suspicious message: Separate a display name from the actual sender, and say what the message alone does not prove. Keep a cloud role to one action and one tenant: Deny a wildcard action, a wildcard tenant, and a principal the role was not written for. See where untrusted text becomes HTML: Point at the HTML context a caller controls, and name escaping or a safe API instead of a homemade word filter. Keep a pull request out of the secret context: Refuse a workflow that checks out fork code in the base repository’s secret context, and pin the action you run. Ask what evidence a control produced: Say what a policy name does not prove, and name one record that would show the control worked. Name what a digital footprint stores: Separate data a person typed from data a service inferred, and say which of those a deletion request can reach. Say who still patches the application: Point at one control the cloud provider runs and one the customer still has to run. Make a detection hit one event: Match the labeled finding and miss the neighboring lines. A rule that matches every row is not a detection. Triage a file before you run it: List what a static look can support, and what it cannot claim without execution. Keep an office change away from a safety system: Name the boundary between an office network and a safety process, and why a convenience change is not automatically safe to copy across.

Open certification prep · Practise the Northbridge case · Open the skill record · Browse labs