Capabilities and Boundaries
The AI Dev Kit equips a general-purpose coding agent with io.Intelligence-specific knowledge and workflows. It supports development decisions and actions, but the agent still works from the project, environment, tools, and permissions available in the current session.
Capability Matrix
| Area | What the kit helps a coding agent do |
|---|---|
| Architecture and package selection | Explain the io.Intelligence topology and choose packages or integration paths for a stated goal. |
| Public API navigation | Find current concepts, APIs, options, examples, and guides in the official documentation and its llms.txt map. |
| Existing projects | Inspect local structure and dependencies, plan an integration, modify code when authorized, and adapt guidance to project conventions. |
| Packaged io.Assist | Materialize or integrate a React or Angular application using published io.Assist packages. |
| Editable io.Assist | Materialize a React or Angular source-owned assistant and guide customization of the one-way fork. |
| Custom AI Web | Materialize the Vanilla JavaScript shell or add AI Web to an existing frontend architecture. |
| MCP | Design, implement, explain, and diagnose tools, resources, prompts, servers, clients, and transports. |
| Working Context | Add or troubleshoot shared business context and connect it to assistant workflows. |
| MCP Apps | Implement or diagnose interactive MCP content, including workspace-oriented integration. |
| Sampling and elicitation | Explain and implement supported human/LLM interaction flows across the applicable packages. |
| Existing agentic backends | Configure AI Mastra Bridge or help implement a custom adapter for the Agent Protocol. |
| io.Assist Anywhere | Guide developers through the official tutorial, answer chapter questions, and diagnose deviations from the guide. |
| Diagnostics | Investigate code, configuration, logs, network flows, and runtime evidence using io.Intelligence-specific playbooks. |
| Template delivery | Retrieve exactly one official template from a catalog-pinned immutable Git tag and record its provenance. |
Backend Responsibility
An Agent Protocol backend is not a prerequisite for installing or using the AI Dev Kit. You can use the kit to learn, plan, inspect code, materialize applications, or develop the protocol adapter itself without one.
An assistant frontend does need a reachable backend before it can run end to end. Building the underlying LLM agentic backend and its business behavior is outside the kit's scope. When a backend already exists, the kit can help expose it through the io.Intelligence Agent Protocol by configuring AI Mastra Bridge or implementing a custom bridge adapter.
V1 Boundaries
V1 intentionally favors a small and transparent development model:
- No version-aware guidance. The kit assumes the installed skill, current official documentation, and locally available io.Intelligence packages are aligned. It doesn't detect versions or select historical API guidance.
- No deterministic project inspector. The coding agent inspects each project with its normal tools and applies the skill's reasoning and diagnostic playbooks.
- No developer MCP server. The initial kit is distributed as one Agent Skill and doesn't require a dedicated MCP service.
- No separate scaffolding CLI or npm product. Template delivery uses a dependency-free script bundled with the skill.
- No bundled template source. The materializer retrieves only the selected template from the immutable Git tag in its catalog.
- No automatic editable-fork upgrades. Editable io.Assist applications become customer-owned source. Future upgrade guidance or diagnostics may be added, but V1 makes no synchronization promise.
- No agent-specific skill variants. One standards-compatible skill is used across Agent Skills-compatible coding agents.
- No universal verification checklist. The coding agent selects validation based on the task and available evidence; the skill constrains unsupported claims.
- No assumed io.Connect runtime proof. Starting a standalone web server or completing a build doesn't establish behavior inside io.Connect Browser or io.Connect Desktop.
- No underlying agent implementation. The kit covers the io.Intelligence protocol boundary around an existing backend, not creation of the LLM agent and its domain logic.
These boundaries describe the V1 contract rather than limitations of io.Intelligence itself. Version selection, editable-fork upgrade assistance, and other support may be considered for later kit releases, but they shouldn't be treated as current capabilities.