Agency Pro GitHub integration for reviewed AEO pull requests

ModelSaid Agency Pro connects an evidence-backed AI visibility finding to a controlled code-review workflow. The system analyzes the connected client repository, proposes a bounded change, validates it, creates a new branch and opens a pull request for normal review. It never pushes directly to the default branch. The agency and client inspect the diff, automated evidence and business facts, then merge or reject it through their established controls. Use Auto Fix for clearly diagnosed repository changes, not for reputation problems, private policy decisions or claims that lack an approved source.

Start with a finding that code can actually fix

A suitable issue has observable evidence, a known public source and a narrow repository target. Examples can include a missing llms.txt file, an unintended robots rule, incomplete visible factual context or structured data that does not match an approved page. A missing independent citation, disputed positioning or stale third-party profile is not solved by changing client code. Confirm the desired business fact with the responsible owner before generating anything. Link the finding to the affected prompt group and define what validation can prove; a clean build proves technical compatibility, not future inclusion in an assistant answer.

• Evidence gate: complete answer capture, reproducible issue and affected prompt group.
• Truth gate: client-approved canonical fact and no invented marketing or regulated claim.
• Scope gate: small set of files, explicit exclusions and no unrelated refactoring.
• Validation gate: repository checks, rendered-content review and artifact-specific validation.
• Review gate: new branch, readable pull request, named approver and deliberate merge decision.

Review the pull request like production work

Inspect the diff for factual accuracy, accessibility, security, framework conventions and unintended page changes. Read generated text in its rendered context and verify that structured data matches visible content with the schema validator. Review crawl directives with the robots.txt generator rather than assuming broader access is always safer. Require ordinary status checks and client-specific reviewers. If validation fails, determine whether the client input or platform caused it under the applicable quota policy. Never bypass protections simply to complete a run.

Operate inside Agency Pro capacity

Agency Pro includes GitHub integration, 15 team seats and 30 on-demand scans per month. Auto Fix is bounded to two runs per UTC day and eight per month for the organization, so it should be scheduled as a high-value implementation resource rather than promised without limit to every client. The plan also includes up to ten businesses, with extra business slots available for purchase, plus the Agency monitoring and white-label capabilities. Confirm current packaging through ModelSaid pricing and write quota allocation into the agency delivery policy.

1. Approve the finding and canonical fact before selecting Auto Fix.
2. Connect the intended repository with least privilege and confirm the correct target context.
3. Review the proposed scope, generated branch, validation evidence and complete diff.
4. Request client or engineering approval through the normal pull-request process.
5. After merge, confirm deployment and later retest the original prompt group without guaranteeing change.

Use a repository readiness checklist before the first run. Confirm the build and test instructions, protected-branch rules, required reviewers, supported package manager, generated-file policy and ownership of deployment. Identify directories that must remain untouched and any security or localization review triggered by public-content changes. A clean connection does not replace this operating context. If the repository cannot validate reliably or the relevant truth has no approved public wording, pause the workflow and resolve that prerequisite before spending an Auto Fix run. Store the checklist with the client workspace. Document repository-specific exceptions and require reapproval when frameworks, build systems or ownership materially change. Keep deployment credentials outside prompts and pull-request descriptions. Revoke obsolete repository access immediately after ownership changes.

Make reviewed code part of the client evidence trail

The agency report should connect baseline answer, diagnosis, approved fact, pull-request URL, validation result, merge decision, deployment and later observation. Begin the wider engagement with the AI visibility scan so code work remains tied to buyer questions rather than generic optimization. Record rejected pull requests and null retests too; they demonstrate that review controls are real. Agency Pro is differentiated by controlled implementation, not autonomous publishing. A bounded new branch and reviewed pull request let technical teams use automation without surrendering responsibility, while the explicit no-default-branch-push rule keeps the final production decision with the client.