Most teams do not fail at security because nobody cares. They fail because security work has no owner, no evidence and no point at which someone must say no. A secure software development lifecycle fixes that by attaching a small number of checks to the moments where work already changes hands: scope approval, design review, merge, release and incident follow-up.
This checklist is written for a product team of five to fifty people, in-house or outsourced, that ships web or mobile software regularly and does not have a dedicated security department. It is a working document. Copy the rows into your issue tracker, name an owner for each, and record what evidence proves the control happened. If a row cannot be evidenced, treat it as not done.
How to use this checklist
- Control: one sentence describing what must be true, not a general aspiration.
- Owner: a role, not a team. "Backend lead" is usable; "engineering" is not.
- Evidence: a link, ticket, report or screenshot a reviewer can open six months later.
- Frequency: per story, per sprint, per release or per quarter.
- Sign-off: the role allowed to accept the risk when the control cannot be met.
Two rules keep the list honest. First, every exception has an expiry date and a named acceptor. Second, the same person cannot both produce and approve the evidence for a high-risk change.
Gate 1: requirements and scope
| Control | Owner | Evidence | When |
|---|---|---|---|
| Classify the data the feature will touch: public, internal, personal, financial or credential | Product owner | Data note attached to the epic | Before design |
| State authentication, authorisation and audit requirements explicitly | Tech lead | Acceptance criteria | Before design |
| Record which third parties receive data and under what agreement | Product owner | Vendor and data-flow note | Before build |
| Define abuse cases for the feature, not only user stories | QA lead | Abuse-case list in the ticket | Before build |
The most common failure here is a feature that collects a new personal field because it is convenient, with no stated purpose and no retention rule. A five-minute question at this gate prevents a year of cleanup.
Gate 2: design and threat modelling
- Draw the data flow: browser, API, queue, database, third-party call and log sink. Mark every trust boundary.
- Ask four questions at each boundary: who can call it, what can they ask for, what is logged, and what happens when it fails.
- Decide where authorisation is enforced. A hidden button is not authorisation; the server must check every object-level request.
- Record secrets handling: where keys live, who can read them, and how they rotate.
- Review file upload, import and export paths, which are common places for unsafe parsing and quiet data leakage.
Evidence: a one-page threat note linked in the design ticket, plus the reviewer's name. Sign-off: tech lead. For features touching payments, identity or sensitive personal data, add a second reviewer who did not write the design.
Gate 3: build and code review
- Branch protection on the main branch, with at least one reviewer who is not the author.
- Automated dependency scanning and secret scanning on every pull request.
- Static analysis rules agreed for the stack, with a documented suppression process.
- Parameterised queries and output encoding as default patterns, not per-developer habits.
- Authorisation tests written alongside the feature, including negative cases.
- No production data in development or test environments unless it has been masked.
Reviewers should be able to answer three questions about any change: what input reaches this code, what privilege does it run with, and what happens to the error. If the answer is that the framework handles it, ask for the test that proves it.
Gate 4: test and release
- Run the security test suite: authentication bypass, object-level access, input validation, file upload and rate limits.
- Confirm the release build contains no debug mode, test keys or sample accounts.
- Check headers, cookies, TLS configuration and error pages against the agreed baseline.
- Review the migration plan: can it be rolled back, and does it expose data during the transition?
- Require a named approver when the change touches authentication, payment or personal data.
Acceptance: the release ticket links to the test run, the approver and any open exception with an expiry date. A green pipeline is not evidence of security testing unless the security tests are part of it.
Gate 5: operate and maintain
- Log authentication events, privilege changes, payment state changes and administrative actions.
- Alert on repeated login failures, unusual export volume and error spikes.
- Track dependency and runtime versions, with a patching window agreed by the business.
- Keep a vulnerability intake path: who receives a report, how it is triaged, and what customers are told.
- Run a short post-incident review after every security-relevant event, including near misses.
Operations is where most teams lose gains made earlier. A patching window that exists only in a document is not a control. Put the dates in the same calendar as the release schedule and name the person who confirms completion.
Verifying the checklist itself
- Pick the most recent release that touched personal or payment data.
- Ask for the evidence for each gate without prompting. Missing evidence is a finding.
- Trace one requirement from the ticket to a test and to the deployed change.
- Check two open exceptions: are they still valid, and did the named acceptor approve them?
- Ask a developer and a reviewer to explain the authorisation model in their own words. Different answers indicate a training or design gap.
Metrics that show whether it works
- Share of releases with complete gate evidence.
- Age of open high-severity findings, measured in days.
- Percentage of dependencies inside the agreed patching window.
- Time from vulnerability report to first response.
- Repeat findings in post-incident reviews, which show whether fixes changed the process.
Common ways the checklist fails
- It becomes a document the security team fills in after the release.
- Every control is marked not applicable without a recorded reason.
- Owners are listed as departments, so nobody is accountable.
- Evidence lives in chat messages that disappear.
- The process is heavier than the change, so teams route around it.
Scale the gates to the risk. A marketing page does not need the same review as a password reset flow. The point is not to slow every release; it is to make sure the changes that could hurt customers receive a different level of attention.
Where our team fits
We apply this gate structure in application and infrastructure work delivered under Managed Technology & Cybersecurity. Engagement scope and starting points are set out on the pricing page. If you are weighing the cost of these controls against their absence, the companion article on secure software development lifecycle cost in Bangladesh explains the cost drivers without quoting a fixed figure.
Sources
- NIST SP 800-218, Secure Software Development Framework, accessed 17 November 2026.
- OWASP Software Assurance Maturity Model, accessed 17 November 2026.
- OWASP Application Security Verification Standard, accessed 17 November 2026.
- OWASP Top 10, accessed 17 November 2026.
Responses (0)
No comments yet. Add the first one below.
Leave a response
Your email address will not be published. Required fields are marked *.