Execution boundary
Determine which pull requests, forks, branches, dependencies and workflow inputs can cause code to execute on privileged infrastructure.
Software delivery chain assurance
A CI/CD pipeline joins source control, workflow logic, third-party actions, runners, credentials, artifact stores, registries and deployment identities into one high-trust system. We model that chain and test whether an untrusted contribution, compromised account or poisoned dependency can cross its approval boundaries.
Decision firstScope the question and relying audience
Safe authorizationSet access, exclusions and stop conditions
Validated evidenceSeparate facts, impact and uncertainty
Owned next moveRemediate, retest or accept explicitly
The buyer decision
Pipeline risk is not limited to a leaked secret in a YAML file. The material question is whether the system lets untrusted code reach a privileged runner, persist in a cache, influence a reusable workflow, obtain a cloud token, alter an artifact or bypass an environment approval. We trace those transitions across the whole delivery chain.
This assessment is useful after rapid DevOps growth, a move to cloud identity federation, adoption of self-hosted runners, new release signing, a supply-chain incident or an enterprise assurance request. Findings name the attacker prerequisite, trust transition, reachable asset and recommended control owner rather than producing a generic hardening list.
Determine which pull requests, forks, branches, dependencies and workflow inputs can cause code to execute on privileged infrastructure.
Validate whether jobs receive only the secrets and cloud permissions required for their specific stage and trust level.
Assess whether builds, packages, images, caches and attestations can be replaced or promoted without the intended review.
Review branch rules, environment approvals, workflow ownership and audit trails for bypass and fragile exceptions.
Assessment coverage
Coverage follows the actual platform—such as GitHub Actions, GitLab CI/CD, Azure Pipelines, Jenkins or another authorized system—without assuming identical control names.
Repository roles, branch protections and review rules
Workflow definitions, reusable components and triggers
Hosted and self-hosted runner isolation
Secrets, variables and cloud workload identity federation
Third-party actions, plugins and dependency trust
Build caches, artifacts, packages and container registries
Environment approvals, deployments and rollback paths
Logging, provenance, signing and incident investigation readiness
Risk paths
OWASP describes poisoned pipeline execution as manipulation of build or deployment configuration to execute malicious commands. We include that path and the surrounding controls that determine its impact.
Assess whether an attacker able to influence source or workflow content can execute in a higher-trust job or runner.
Test whether fork or pull-request events receive secrets, writable tokens, caches or access to persistent infrastructure.
Review mutable references, reusable workflow trust, plugin updates and repository transfers for execution hijack.
Evaluate isolation, cleanup and network reachability where a job could affect later jobs or internal systems.
Map OIDC claims, audience, subject conditions and role permissions from pipeline job to cloud impact.
Challenge upload, promotion, registry and deployment controls that bind reviewed source to released output.
Look for alternate branches, manual jobs, inherited permissions or administrators that circumvent intended separation.
Assess log retention and audit coverage for workflow changes, secret access, artifact mutation and deployment actions.
Scope architecture
The resulting model makes remediation easier because owners can see which single control breaks an entire attack chain.
| Trust transition | Representative questions | Primary evidence |
|---|---|---|
| Contributor to workflow | Can a fork, pull request or low-privilege maintainer change code that executes with repository or organization trust? | Trigger logic, workflow ownership and branch rules |
| Workflow to runner | Does untrusted code reach a persistent or network-privileged runner, and is state removed between jobs? | Runner groups, labels, network paths and cleanup model |
| Job to credential | Are repository tokens, secrets and federated cloud roles constrained by event, environment, branch and job? | Permissions, secret scopes and identity claims |
| Build to artifact | How is a reviewed commit bound to packages, images, signatures and promotion? | Artifact metadata, registry roles and release controls |
| Artifact to production | Can deployment approval, environment protection or rollback be bypassed or invoked with altered output? | Environment rules, deploy identities and audit events |
Delivery process
Every phase preserves scope, authorization, evidence quality and a practical path to closure.
Inventory repositories, triggers, shared workflows, runners, credentials, artifact stores and deployment environments. Classify who can influence each stage.
Inspect approved workflow and platform evidence for dangerous triggers, excessive permissions, mutable dependencies and control exceptions.
Use test repositories, non-production environments or inert proof where possible. Never publish or deploy malicious output merely to demonstrate theoretical impact.
Prioritize changes that remove high-trust execution, narrow credentials, isolate runners and bind artifacts to approved source. Provide retest cases.
Deliverables
Deliverables are tailored to the agreed audience. Technical evidence remains access-controlled; summaries keep the tested scope, date and limitations attached.
CI/CD architecture and trust-transition map
Repository, workflow, runner and identity inventory
Validated attack paths with safe evidence
Secrets and permission exposure analysis
Artifact integrity and promotion observations
Platform-specific remediation actions
Control-owner and priority mapping
Retest results for material corrections
Service fit
Evidence boundaries
We agree in advance whether validation uses a test repository, disposable runner, non-production environment or an inert production-safe proof. The standard assessment does not deploy malicious code, alter a customer release, extract unnecessary secrets or weaken branch controls. Any exception needs explicit authorization and rollback planning.
OWASP CI/CD guidance and the NIST Secure Software Development Framework inform coverage, but the engagement is neither an OWASP certification nor an SSDF conformity assessment. The report reflects the repositories, workflows, runners, identities and artifact paths available during the agreed window.
Common questions
We scope around the delivery architecture rather than one brand. Common environments include GitHub Actions, GitLab CI/CD, Azure Pipelines and Jenkins, along with artifact registries and cloud deployment identities. Capability and required access are confirmed before commercial commitment.
No, not as a default method. We prefer test repositories, isolated runners, non-production stages or inert evidence. Any validation that could change an artifact or deployment requires explicit authorization, defined rollback, monitoring and a clear reason the safer alternative is insufficient.
Yes, when included in scope. We examine where secrets are available, how tokens are minted, which claims constrain workload identity and what cloud permissions the job receives. Sensitive values are minimized in evidence and should never be sent through the public form.
No. Source code review focuses on the security of application implementation. A CI/CD assessment focuses on the trust system that builds, signs, stores and deploys that code. The two can be combined when custom pipeline tooling or release-critical code also needs review.
Share the pipeline platforms, approximate repository count, runner model, artifact and registry services, deployment clouds, use of reusable workflows and decision deadline. A high-level diagram is helpful; detailed configuration and credentials can wait for NDA and secure evidence transfer.
Related services
Next step
Share the decision, systems, important roles or environments, access constraints and deadline. Begin without credentials or protected evidence; NDA and secure transfer can be agreed before sensitive material is exchanged.