pcb-project-intake

SkillDocs & knowledge

RAG-backed. Inventory a bounded DipTrace project, its document identity, models, rules, connectivity, unknowns, and evidence before planning changes. Use when the user says “Inventory this DipTrace project before we plan work.”

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the pcb-project-intake skill

What this skill tells your AI

The instructions your AI receives, as published by fireostendere/mcp_diptrace in skills/pcb-project-intake/SKILL.md and read by ahel’s review.

Read runtime access before choosing between explicit-path MCP, a live bridge session, and native/headless CLI. Their availability is separate.

PCB project intake

RAG: engineering memory by defaultshared workflow. Build project context from recorded requirements, architecture and prior lessons; relate the exact file/model inventory to design intent and the next engineering decisions.

Produce a read-only project baseline. Do not turn absent fields into assumed requirements. Use public tools/list for exact callable names and get_capabilities for document/configured feature availability.

Workflow

  1. Call diptrace_status, then get_capabilities.
  2. Resolve each document as XML or a native binary before get_document_info. Explicit-path XML inspection does not require a live editor. For a binary, find its corresponding XML export or use the supported native workflow on a copy; base roundtrip alone does not export XML. Record kind, literal format version, document units, byte size, and SHA-256. Treat PCB, schematic, component library, and pattern library as different source types.
  3. Read only advertised models: get_board_model, get_schematic_model, get_design_rules, and get_connectivity_graph. Use count-only or paginated board reads when the response is large.
  4. Separate caller requirements from document facts. Record an unresolved requirement as blocked_by_input only when it changes topology, ratings, pin mapping, mechanical datum, units, or release criteria.
  5. Return the inventory through ../shared/result.schema.json.

Quantitative boundaries

  • All MCP geometry is millimetres even when root Units is inch or mil; preserve the literal root value as document evidence.
  • get_board_model accepts page limits from 1 through 500 and has a count-only mode; use the advertised byte budget instead of requesting the whole board.
  • One write transaction is limited to 100 staged operations and 500 conservatively counted affected objects/elements; the count may exceed unique physical objects. This skill never opens one.
  • A SHA-256 is exactly 64 lowercase hexadecimal characters. Never substitute a path or timestamp for document identity.

These limits come from server.py, capability_model.py, write_limits.py, and runtime get_capabilities; runtime values win.

Refusals

  • Do not infer DipTrace 5.3 compatibility from a synthetic scaffold or parser success.
  • Do not call a tool absent from runtime discovery; follow ../capability-map.json.
  • Do not collapse schematic logical connectivity and PCB ratlines into one completion claim.
  • Keep source designs unchanged. Native checks requested as part of intake use isolated copies; discover the helper before calling native opening unavailable. Do not start unrelated processes.

Label every returned fact caller or document; label derived prioritization heuristic.

Signals

GitHub stars
22
Forks
1
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
pcb-project-intake
Source
github.com/fireostendere/mcp_diptrace