Dependency scanning
Known vulnerabilities in third-party packages surfaced at pull request time, with an upgrade path attached.
Security findings delivered as a report after the fact get triaged, deferred and forgotten. The same findings delivered as a failed build get fixed that afternoon, because the change is still in the author's head and nothing ships until it is resolved.
Fig. 01Pipeline security checksIllustrative
Typical stack
The aim is not more scanning. It is scanning positioned where the result is actionable, with thresholds agreed in advance, so the team is never negotiating severity in the middle of a release.
Agreed thresholds. What blocks and what warns is decided up front, so a gate is never argued about under deadline.
Secrets out of the repository. Detection in the pipeline and in history, with a managed secret store as the replacement.
Policy as code. Infrastructure rules — public buckets, open security groups, unencrypted volumes — enforced automatically.
Least privilege by default. Scoped service accounts and short-lived credentials rather than shared long-lived keys.
Known vulnerabilities in third-party packages surfaced at pull request time, with an upgrade path attached.
Base image and layer analysis on every build, with a policy on which severities block a release.
Pre-commit hooks and pipeline scanning across the working tree and history, with a rotation procedure when something is found.
Code-level security checks tuned to the actual risk, because a noisy scanner is one the team learns to ignore.
Terraform plans evaluated against policy before apply, so a misconfiguration never reaches the account.
Least-privilege roles, short-lived credentials and audit logging of privileged actions, retained where you can query it.
Each control is placed at the earliest point where its result is actionable. A finding at deploy time is expensive; the same finding at pull request time is routine.
Every engagement ends with artefacts you own outright and can hand to another team without calling us first.
Scanning wired into the pipeline. With the blocking thresholds documented and agreed rather than left at a tool's defaults.
A secret management setup. Secrets in a managed store, removed from the repository, with rotation documented.
Policy-as-code rules. Version-controlled infrastructure policies enforced before apply.
An access review. Current permissions mapped, over-broad grants identified, and a least-privilege target documented.
Scanning adds time, so it is placed where it is cheapest — incremental checks on pull requests, full scans on the main branch. The goal is that nobody has a reason to argue for turning it off.
They get triaged into a prioritised backlog. Setting the gate to block everything on day one stops all delivery and gets switched off within a week; the gate starts at agreed severities and tightens over time.
It builds the technical controls auditors look for — access control, change management, audit logging, vulnerability management. The audit itself is between you and an accredited firm; we are not auditors and do not issue certifications.
No. We build the controls and remediate findings. Independent penetration testing should come from a party that did not build the system.
If the honest answer is “by whoever happened to notice”, that is the place to start. Tell us how your pipeline works now.