Start in your repository.

Work in your own repository

This repository supplies the tools. Keep your application and its architecture in a separate repository. The plugin supports new applications and AI additions to existing systems.

For scope and examples, read What can you build?. If you already have an application, prepare its context before discovery.

Before scheduling the working session, check the participant requirements. The business participant does not need every technical permission personally, but the customer must name the specialists and approvers who can resolve access and architecture questions.

For a new project, create an empty directory and initialize Git yourself:

mkdir my-ai-app
cd my-ai-app
git init

For an existing application, clone or open its repository instead. Follow its branching and approval rules. Even a new project may need to use existing enterprise systems and controls. Tell the creator which ones it must reuse.

Open your application repository in VS Code, or start Copilot CLI there. Use an approved workspace and synthetic inputs until you have agreed on customer-data access.

Install the plugin

Choose your host. Review the plugin contents and publisher before granting trust. Install the supporting skills separately in your project.

GitHub Copilot CLI

Run these commands in a terminal:

copilot plugin marketplace add microsoft/frontier-agentic-cobuild-rvas
copilot plugin install agentic-cobuild@agentic-cobuild
copilot plugin list

The native CLI installation is user-scoped. It makes the plugin available across repositories; it does not copy application code into your workspace.

VS Code Copilot

In your user settings, enable chat.plugins.enabled and add this repository to the existing marketplace list without replacing other entries:

{
  "chat.plugins.enabled": true,
  "chat.plugins.marketplaces": [
    "microsoft/frontier-agentic-cobuild-rvas"
  ]
}

Open the Extensions view, search for @agentPlugins, and install Agentic Co-build. Review the marketplace trust prompt. You can also use Chat: Open Customizations to manage plugins. Enable it for your application workspace.

VS Code can discover CLI-installed plugins when both hosts share the same user environment. Remote, container, or WSL environments may use different locations. Check availability in the environment where you will build; do not assume a Windows installation is available inside a container.

The plugin adds AI Agent Creator, Idea Forge, and Cloud Architecture Diagram. It also supplies the public Microsoft Learn MCP server. This server retrieves documentation; it cannot read or change your cloud resources. No live Azure, Foundry, or hosted diagram connection is installed by default.

Install the supporting skills

Use Node.js 22.20 or newer, Git, and npx. Run the commands below from your application directory. You do not need Bash or a clone of this tooling repository.

Each command installs the listed skills directly from its upstream repository. Review the skills and their source before running it. Check .agents/skills/ and .github/skills/ for existing copies and preserve any local changes.

The commands select GitHub Copilot at project scope. --copy copies skill files instead of creating symlinks. The first --yes lets npx download the pinned installer; the final --yes accepts the skill installer's prompts. These commands can overwrite existing copies of the selected skills.

The installer version is pinned, but upstream skill content is not pinned to specific commits. Review upstream changes before installing or reinstalling. Keep skills-lock.json and installed skill files according to your project's source-control policy.

Skills from mattpocock/skills

Upstream source: mattpocock/skills.

npx --yes skills@1.7.1 add mattpocock/skills --skill grilling domain-modeling codebase-design implement tdd diagnosing-bugs code-review handoff --agent github-copilot --copy --yes

Skills from microsoft/azure-skills

Upstream source: microsoft/azure-skills.

npx --yes skills@1.7.1 add microsoft/azure-skills --skill discover-azure-skills azure-enterprise-infra-planner microsoft-foundry azure-ai foundry-iq entra-agent-id entra-app-registration azure-aigateway --agent github-copilot --copy --yes

If a command fails, stop and inspect its error, .agents/skills/, and skills-lock.json before retrying. A failed install may leave partial files. Preserve existing skills rather than deleting them to bypass a conflict.

Third-party skills remain upstream-owned. Installing a skill does not authorize cloud changes. The upstream implement skill includes a commit step. Follow the implementation handoff below and authorize commits separately.

Check the workspace before starting

In CLI, start a fresh session in the application directory. Use /skills and /mcp to inspect available skills and the Microsoft Learn connection.

In VS Code, use Chat: Configure Skills and the MCP server list. Confirm the plugin is enabled, the supporting skills are discovered, and Microsoft Learn can retrieve documentation. Your host may show a prefix on plugin skill names. Select the exact entry in its picker or slash completion.

If skills are missing, check the current workspace, plugin enablement, host version, and organization policy. Restart or reload after installation. If the documentation server is unavailable, report the limitation; do not invent current product facts.

Python 3.10 or newer is required for local diagram generation. Local draw.io rendering is a separate, organization-approved prerequisite. The plugin does not install a renderer or upload your architecture to a hosted service. If visual or independent review is unavailable, record the gap. You must explicitly accept a manual-review exception before architecture approval.

Run an initial discovery session

If the idea is distributed across customer participants, run a recorded discussion before starting AI Agent Creator. The discovery session guide provides the facilitator's questions, a discovery-brief structure, and handoff prompts for WorkIQ or an approved transcript.

The customer must approve recording and AI processing under its policy. Review the resulting brief with the customer. The brief is discovery evidence, not requirements approval or permission to start architecture work.

Describe what you want to build

Select AI Agent Creator (ai-agent-creator) and give it the intended outcome. You do not need to choose the platform or runtime first.

For a new application:

/ai-agent-creator Design an AI maintenance assistant for property managers that triages
resident requests and prepares work orders for approval. Account for our
existing enterprise identity and operating environment.

For an existing application:

/ai-agent-creator Add evidence-backed report summarization to this existing application.
Inspect its current stack, preserve established interfaces, and account
for our enterprise data-access and deployment constraints.

The creator inspects the repository and interviews you. Confirm discovery once you agree on the outcome and first user journey. Settle the decisions that affect the architecture before confirming. The creator then prepares:

Review the complete package, including any recorded review exception. Approval ends this workflow. It does not start implementation, change application code, generate infrastructure, or provision resources.

Start implementation in a new session

Choose a model suited to sustained engineering work. Attach the current implementation-plan.md, specification.md, and solution.md from your application repository. Start with this prompt:

Implement the attached plan. Treat the attached specification as the
requirements source of truth and follow the attached architecture.
Work in dependency order and validate against the acceptance criteria.
Report each blocked criterion with its reason and owner.
Ask before commits, resource provisioning, or deployment.
Do not treat implementation as complete until every applicable acceptance
criterion is verified or explicitly reported as blocked.

Use the application's existing tests and delivery controls. If scope or architecture changes an approved decision, update the package and get that decision approved again before continuing.

Review before deployment

After you finish implementation and run its targeted checks, run this in CLI:

/review Review the staged and unstaged implementation against
@docs/architecture/specification.md,
@docs/architecture/implementation-plan.md, and
@docs/architecture/solution.md. Identify unmet or partially met acceptance
criteria, skipped or out-of-order plan steps, architecture deviations,
regressions, and missing tests. Rank findings by severity and include file and
line references.

Resolve the findings and rerun the affected checks. Then run a separate security review:

/security-review Analyze the staged and unstaged implementation for exploitable
security vulnerabilities, insecure defaults, authorization or data-isolation
failures, secret exposure, injection risks, unsafe external interactions, and
dependency or configuration weaknesses. Rank findings by severity and confidence,
include file and line references, and recommend the smallest safe fix for each
finding.

Resolve the reported security findings and rerun the affected tests before approving deployment.

In VS Code, request a separate implementation review against those same files. Then request a separate security review using the review tools available in your organization. Do not assume CLI slash commands exist in VS Code. If independent review tooling is unavailable, assign a human reviewer and record the gap.

Setup and architecture approval are not production-readiness checks. The operating team must accept the deployed user journey and take responsibility for rollback.

Need an idea first?

Select Idea Forge (idea-forge) in the same application workspace. Provide a customer name and industry, with any known pain points:

/idea-forge Find a bounded AI opportunity for [customer] in [industry].
Research the business and publicly stated priorities on the web.

Idea Forge researches the customer's business, industry, and publicly stated priorities. It returns separate internal and external shortlists, each with up to five ideas: two Simple, two Medium, and one Complex. If public evidence cannot support that mix, it explains the gaps.

Choose an idea to receive a brief with evidence and open questions. The brief is not yet approved. Invoke AI Agent Creator separately with it for detailed discovery and architecture approval.

Updates and team setup

Each teammate installs the plugin in their own host and checks the project skills. Use CLI's copilot plugin update agentic-cobuild or VS Code's plugin update action for plugin updates. Review the updated contents before use.

Supporting skill updates are separate. Review source changes and your project's lock file before reinstalling selected skills with the commands above. Preserve local changes and inspect the resulting diff before committing.

See the plugin source and the diagram asset terms.