What problem does it solve? Creating custom agents by hand leads to over-privileged capabilities, inconsistent definitions, and unreviewed changes to .agent.md files. This Skill turns an agent idea into a validated, least-privilege blueprint with deterministic rendering, review evidence, rollback-safe installation, and governed publication handoffs for Foundry endpoints, Microsoft 365 Copilot and Teams, or ChatGPT actions. ## Core Features & Use Cases - Blueprint-driven agent authoring: Define role, capabilities, autonomy mode, approval gates, subagents, and handoffs in a schema-validated JSON blueprint, then render a deterministic .agent.md file. - Transactional installation: Plan, review, and apply agent changes atomically with hash verification, symlink protection, and automatic restore from backup if verification fails. - Publication handoff planning: Record schema 2.2 publication intent for Foundry endpoints, Microsoft 365 Copilot and Teams, or ChatGPT actions without creating any external resource. - Use Case: Build a read-only accessibility reviewer agent with only read and search capabilities, preview the rendered agent and plan, obtain explicit risk acceptance, then install it under .github/agents/ with a verified hash. ## Quick Start Ask the agent-builder skill to create a read-only code review agent with read and search capabilities and preview the plan before installing it.
