Strategy and approach for security review across the pipeline
Security Modernization
Every artifact that passes through the Forge pipeline is inspected by a dedicated Security Agent. Not a post-hoc scan: security analysis happens at generation time, before the artifact reaches you.
What the Security Agent actually checks
Five named checks run against every artifact as it is generated, plus a set of guards specific to how AI systems get attacked. All of it is visible before you approve a stage.
Auth coverage
Is authentication defined for every user-facing endpoint? Gaps surface while the architecture is still a document, not after the endpoint ships.
Secrets handling
No hardcoded credentials in specs or in generated code. Detected secrets are also redacted from connector-backed context before they are persisted or displayed.
BEHAVIOR PRESERVED
Input validation
Every user input carries a validation requirement, checked as a property of the specification rather than left for a code reviewer to notice.
Dependency risk
Third-party libraries are flagged for known CVEs, and supply chain dependency risk is assessed while the architecture is being reviewed.
Compliance gaps
Missing controls for the compliance requirements you declared. Requirements captured at intent are carried forward rather than rediscovered late.
Guards built for AI-generated systems
Prompt injection patterns in specs, MCP tool definitions that could leak context to external endpoints, and scope creep beyond the approved specification boundary.
See your ForgeScore before you commit
Every import is scored across eight engineering-health dimensions before a Work Order is planned, so you know what you are dealing with up front.
Security review on Forge is not a stage you reach at the end. A Security Agent inspects intent, the PRD, the architecture, and the user stories as each one is generated, and ForgeScore baselines the codebase you started from.
8
ForgeScore dimensions
Every stage
Approval gates
6
Pipeline stages
Where security shows up in the pipeline
The shift is more than syntax. Paradigms, data access, and runtime assumptions all move, and each one needs an explicit mapping.
01
At the intent stage
Ambiguous or conflicting requirements that could lead to insecure designs, missing security requirements such as auth, data protection and audit logging, and compliance requirements that must be carried forward.
02
At the PRD stage
Features that require security controls are flagged for explicit inclusion in the architecture, alongside out-of-scope exclusions that might create security gaps and data classification requirements.
03
At the architecture stage
OWASP Top 10 mitigations present in the design, authentication and authorization flow completeness, secrets management patterns, supply chain dependency risks, and network segmentation with least-privilege service design.
04
At the user stories stage
Security stories explicitly included in epics, penetration testing tasks included, and dependency scanning tasks included, so the work is scheduled rather than assumed.
What Forge actually produces
Every finding carries severity, the file it was found in, an effort estimate, and its compliance references. Here is one Trust Boundaries finding as it appears in the report.
ForgeScore dimension
Trust Boundaries
42
Security boundaries between components and data flows. Foundation maturity: auth patterns exist and are well built, but enforcement is inconsistent across routes.
High severity finding
Trust Boundaries finding
Service-role key used as silent fallback when anon key would suffice
OWASP-A01
One route falls back to the anon key when the service key is missing, while others default the opposite way. A misconfigured environment silently escalates or de-escalates privilege. Tagged OWASP-A01 and CWE-269, medium effort, high velocity impact, with a suggested fix attached.
Common challenges
Security review surfaces the decisions a project tends to defer. These are the ones worth resolving before you approve a stage rather than after.
Findings do not block progression
By design, findings are informational. That keeps the decision with your team, but it also means someone has to own triage. Without that, an amber callout is just a callout.
Declared compliance drives the gap check
The Compliance Gaps check looks for missing controls against the requirements you declared. If the requirement was never captured at intent, there is nothing to check against.
Prompt injection is a spec-level risk
Forge detects patterns in specs that could enable user-controlled prompt manipulation. That is a class of problem most review checklists were never written to catch.
Tool definitions can leak context
MCP tool definitions that could exfiltrate context to external endpoints are flagged. If you are connecting agents to internal systems, this is the exposure to understand first.
Frequently asked questions
Is this a vulnerability scanner?
Does this cover the OWASP Top 10?
Do security findings block the pipeline?
What are the eight ForgeScore dimensions?
How do I prove security improved?