implement-guardrails-accountability

GitHub Certified: Agentic AI Developer (GH-600) · 22 questions

• A GitHub Copilot agent can perform the following actions: (1) read repository files, (2) create a branch, (3) deploy to production, and (4) write a new secret to GitHub Secrets. How should these actions be classified by risk level?
• A GitHub Copilot agent is attempting to perform a force push to the main branch, which violates the organization's branch protection policy. How should the guardrail system respond?
• A developer is designing the guardrail policy for a deployment agent that can perform four actions: (1) read repository files, (2) create feature branches, (3) force-push to the main branch, and (4) deploy to the production environment. How should these four actions be classified by risk level for the purpose of autonomy assignment?
• A team is designing the autonomy policy for their CI/CD agent. They are considering three actions and two proposed assignments: (A) branch creation — fully autonomous, (B) pull request merge — human-in-the-loop review required, (C) production deployment — explicit human approval required before action. Which statement BEST describes this autonomy assignment?
• An organization's policy requires all code changes to the main branch to go through a pull request with at least one human approval before merging. A code generation agent attempts to push commits directly to the main branch, bypassing the pull request process entirely. Which GitHub guardrail configuration BEST prevents this action?
• A pull request comment agent has two required capabilities: (1) read repository file contents, and (2) create and update comments on pull requests. The agent is currently configured with a GitHub personal access token (PAT) that grants full repository admin access, including settings, webhooks, branch protection management, and member administration. What change should be made to align the agent with least privilege principles?
• A cleanup agent is configured to remove unused GitHub Actions environments and their associated secrets as part of a repository maintenance task. The agent has determined that an environment named staging-legacy qualifies for deletion. Deleting this environment and its secrets is permanent — the secrets cannot be recovered after deletion. What guardrail MUST be in place before the agent executes this deletion?
• A developer discovers that an agent has been merging pull requests directly to main without any human review. The branch protection rule requiring 'at least one approving review' is configured on the repository, but the agent's GitHub App has the 'bypass branch protection' permission enabled. No merge failures have occurred. What is the guardrail failure?
• A security audit reveals that an agent tasked with generating configuration files has committed API keys, database connection strings with passwords, and private key material directly into a public repository. The agent had no secret detection configured and treated configuration values the same as any other file content. What guardrail was missing and what is the remediation?
• A team reports that their database maintenance agent deleted a production database during a cleanup operation. The agent had been given broad database management permissions and interpreted a 'remove old test databases' instruction as applying to any database matching certain naming patterns. The production database matched the pattern. No authorization was requested before the deletion. What guardrail design failure does this most directly represent?
• A security audit of your deployment pipeline agent reveals that the agent has been granted write access to all 47 repositories in the organization, admin access to CI/CD pipeline settings, and the ability to modify branch protection rules. However, the agent's actual task is only to read build logs and generate a daily summary report. What security principle has been violated, and what is the recommended remediation?
• A financial services company sets an agent's autonomy level to 'full auto' for all operations, including generating and submitting regulatory compliance reports to external authorities. During an audit, it is discovered that several reports were submitted with errors, and the company cannot demonstrate that any human reviewed and authorized these submissions. Which two issues does this create?
• Your organization defines three agent autonomy tiers: Tier 1 (read-only exploration), Tier 2 (can propose changes via PR), Tier 3 (can merge approved PRs). Which GitHub Actions workflow permissions mapping is correct for this tiered model?
• An organization wants to ensure that no agent-produced code reaches the main branch without passing a dependency vulnerability scan. The scan must be mandatory and cannot be bypassed by the agent or any developer without administrator override. How do you configure this guardrail?
• A deployment orchestrator agent determines that a hotfix is ready and instructs a deployment worker agent to deploy directly to the production environment. The worker agent has the technical capability to execute the deployment. No human has been notified. Which accountability principle is violated, and what is the correct design?
• A coding agent needs to create issues and push commits in three specific repositories, but should have no access to any other repositories in the organization. What is the most secure authentication configuration for this agent?
• Your organization requires that every AI agent's autonomy level (and the permissions granted to support it) be formally documented with a rationale that can be audited during a compliance review. Where and how should this documentation be created?
• An organization is designing a graduated autonomy program for its fleet of 12 agents. They want a framework that progressively grants agents higher autonomy levels as trust is established. Which element is MOST essential to include in this program to prevent unsafe autonomy expansion?
• A DevOps team is deploying an agent that manages Kubernetes cluster resources. The team is determining the appropriate autonomy level for the agent. The agent can: scale deployments up or down based on metrics, evict pods when nodes are under memory pressure, update ConfigMaps with new application settings, drain nodes before maintenance, and delete PersistentVolumeClaims (PVCs). The team has four autonomy levels available: (1) Suggest-only, (2) Suggest-and-execute with human confirmation, (3) Execute-and-notify, and (4) Fully autonomous. Which autonomy level assignment across these actions BEST reflects appropriate risk-proportionate governance?
• A compliance officer is reviewing an agent's workflow for processing employee termination requests. The workflow includes: (1) receiving the termination request from HR, (2) disabling the employee's SSO account, (3) revoking all active OAuth tokens, (4) archiving the employee's email to a compliance mailbox, (5) removing the employee from all GitHub teams and organizations, (6) transferring repository ownership for any repos the employee owns, and (7) deleting the employee's corporate laptop MDM enrollment. Which steps in this workflow MOST require mandatory human judgment gates, and why?
• A Copilot agent with production deployment access is about to delete a database table as part of a cleanup task. This action cannot be reversed without a backup restore. What authorization model best governs this specific action?
• An enterprise's agentic deployment system requires human approval for every action. After three months, the SRE team reports severe 'approval fatigue' — reviewers are rubber-stamping requests without real review because they receive 200+ approval requests per day, mostly for low-risk routine operations. This is paradoxically creating more risk than a better-designed autonomous system would. Which design approach would MOST effectively reduce approval overhead while maintaining meaningful human oversight for genuinely risky actions?