AXLEKEYSBETAPhysical objects need more than a good prompt. They need context, rules, parts, dimensions, validation, and review. Agents and Skills help Axle Keys manage that complexity through clear roles, reusable knowledge, and safe workflows.
A single assistant trying to plan, design, code, validate, and repair all at once often becomes unreliable. Axle Keys separates the work into agent roles — each with a clearer responsibility.
What the system has to understand
The safest agent is not the most autonomous one. The safest agent knows its job, uses the right context, and makes changes in reviewable steps.
Not every request needs every role. A parameter change may only touch three. A new model may need the full chain. A failed model routes directly to repair.
Always-on coordinator. Understands the request, checks project context, chooses the workflow, loads the right skills, routes tasks to the right roles, and keeps the user informed. Prevents agents from guessing when context already exists.
Translates the user's request into a practical modeling plan before code is written. Defines object type, System selection, main parts, spatial relationships, modeling approach, build sequence, and areas that need user confirmation.
Decides which dimensions and options become editable parameters. Defines labels, safe ranges, default values, and hidden internal values. Goal: useful control, not endless control — protect the model from confusing or invalid changes.
Creates or updates CAD model code using Replicad and System-specific modeling patterns. Should read the existing model before changing it, follow the Design Planner's plan, preserve working code, and make focused changes. Executes the plan — does not guess the whole design.
Checks whether the model built correctly and the output appears usable. Covers build status, syntax errors, geometry errors, missing parts, invalid outputs, warning flags, and System-specific checks. Not a fabrication guarantee — a structured error check.
Acts when something fails. Job is not to redesign the object — it is to identify the failure and make the smallest useful repair. Reads the validation error, inspects the current model, preserves working parts, creates a focused fix, and re-runs validation.
Converts images, sketches, screenshots, or visual references into useful project context — object type, proportions, visible parts, style cues, layout, likely dimensions, and areas that need confirmation. Useful when the user uploads reference images.
Reads technical documents, drawings, manuals, diagrams, specifications, or CAD-related reference material — extracting dimensions, constraints, material references, part names, hardware references, tolerances, and warnings.
Turns written briefs, notes, emails, client descriptions, or project instructions into structured requirements — goals, constraints, object types, dimensions, user preferences, missing information, and action items.
Understands panels, openings, shelves, doors, drawer fronts, material thickness, backs, toe kicks, edge banding, and cut lists — helps the workflow stay inside Cabinet Systems logic.
Understands profiles, cut lengths, brackets, fasteners, panels, plates, clearances, and modular assembly logic — helps the workflow stay inside T-Slot Systems logic.
A skill is a reusable packet of knowledge, instructions, rules, examples, or tool guidance. Without skills, agents guess. With skills, agents follow established patterns.
The agent is the role
The skill is the knowledge
The tool is the action
The project is the context
Skills make the workflow consistent: same startup process, same routing logic, same validation approach, same System rules, same repair process. A skill is not just documentation for humans — it is operational context for agents.
Entry point for an agent session. Defines what to do at startup: identify the active model, load core pipeline skills, load System-specific skills, read existing model code before changing it, choose a workflow route, use drafts for meaningful changes, validate before promotion.
Define how each core role works — Project Manager, Design Planner, Parameter Designer, Code Agent, Validator, Repair Agent. Each describes responsibilities, inputs, outputs, rules, and failure behavior. Keeps the agent workflow predictable.
Describe a way of building — Open Design, Cabinet Systems, T-Slot Systems. Include available parts, parameter rules, modeling assumptions, material defaults, output expectations, constraints, examples, and validation checks. What makes agents more useful than generic CAD helpers.
Explain how to use specific software tools — Replicad API reference, Three.js view rules, file/model/draft/validation/export tools, MCP tool descriptions. Help agents use tools correctly instead of inventing unsupported actions.
Working patterns — a simple cabinet, a T-slot frame, a closet bay, a working parameter setup, a known repair pattern. Agents often perform better when they can follow a known pattern.
Axle Keys works with physical objects. Agent output should be reviewed before it becomes important.
The safe pattern
When approval matters most
What each step means
Different tasks route through different agent combinations. Simple changes stay shallow. Complex tasks use the full pipeline.
Should not require a full model rewrite.
Repair should be narrow.
The image becomes project context, not just inspiration.
MCP is how agents call Axle Keys tools. Agents & Skills describes how work is organised. Together they form the intelligence layer above Studio.
Related pages
Clear roles, reusable knowledge, validation, and human review make agent-assisted CAD more reliable — not less. The point is not to remove responsibility. The point is to make the work easier to inspect.