Best practices quick reference
The highest-impact recommendations from this guide, without the reasoning behind them. Each one links to the section that explains why.
Source and setup
| Do this | Why it matters | Details |
|---|---|---|
| Start with one application you know well | The first scan of a large codebase can produce hundreds of issues and pull requests. | Prepare and connect source code |
| Connect source, not compiled output | A ZIP with only binaries failed at the scan stage. | Check source eligibility |
| Keep Amazon Inspector for packages and binaries | Continuum flags a library only when its flaw is reachable from application code. | Where Continuum fits |
| Use private repositories | Each fix opens a pull request that describes a vulnerability. | Connect the source |
| Include deploy workflows and infrastructure-as-code in the scanned source | They let Continuum relate a flaw to a running workload. | Prepare the build inputs |
AWS context
| Do this | Why it matters | Details |
|---|---|---|
| Create environments in one central account and put a collector in each workload account | One place for issues and patches, and each account owner keeps control of its inventory. | Use one central environment |
| Give the environment's data access role access to every associated collector | Associating a collector with a new role replaces the stored role. | Configure the data access role |
| Read likelihood and impact factors, not only severity | They show whether a rating reflects your deployment or only the code. | Add EKS and deployment context |
| Write a short business-context document for S3 sources | It is where your judgment about sensitive data enters a rating. | Identify your crown jewels |
Scans, issues, and patches
| Do this | Why it matters | Details |
|---|---|---|
| Scan when code changes, and triage by flaw | Rescans of unchanged code can add new issues with new IDs. | Scan and remediate |
| Wait for issue processing before concluding anything | A scan can succeed while issues and patches are still arriving. | Follow issue processing |
| Describe traced issues as traced | Every issue in our runs was proved by source analysis. | Read the evidence |
| Treat every patch as a draft | A patch is a proposal, and delivery is not deployment. | Review a patch |
| Use custom findings to check another tool, and keep that tool | Continuum confirmed and re-rated most findings but added little new detection. | Submit findings from another source |
Cost
| Do this | Why it matters | Details |
|---|---|---|
| Set an issues-per-scan limit and a focus mode that fit your budget | Findings that are not pursued are not validated, rated, or patched, so a cap makes the bill predictable. | Limit what Continuum works on |
| Set a budget alert before the first scan | The first scan starts at creation and spend is not capped. | Cost considerations |
| Match the schedule to how often you can triage | Issue count drives inference charges. | Control cost |