CAD DWG/DXF Drawing Analysis Tool

Purpose

Provide professional-grade DWG/DXF drawing file reading and analysis capabilities. This skill enables structured data extraction from CAD drawings, spatial analysis between entities, visual capture of drawing regions, and standards compliance auditing.

⚠️ Security Notice

Before using this skill, please review the following:

This skill does NOT automatically install any packages or run any setup scripts. If dependencies are missing, it will output a clear error message with manual installation instructions. To set up the environment, the user must explicitly run the setup command:

CODEBLOCK0

The full setup process (when explicitly confirmed) will:

• - Install Python packages (ezdxf, matplotlib)
• Install system packages (xvfb, libGL) via INLINECODE4
• Download and install ODA File Converter from INLINECODE5
• Download and install QCAD dwg2bmp from INLINECODE6

All downloads are from official sources, but the skill requires network access and elevated permissions.

Prerequisites

• - Python 3.8+ (system installed)
• Python packages: ezdxf and matplotlib must be installed before use
• root/sudo permissions (only if you choose to run automatic setup)
• x86_64 Linux system

Setup Options

Option A: Manual Setup (Recommended for security-conscious users)

Install dependencies yourself before using the skill:

CODEBLOCK1

Option B: Assisted Setup (Requires confirmation)

Run the setup script with explicit confirmation:

CODEBLOCK2

What setup does (with your confirmation):

1. 1. Installs Python dependencies (ezdxf, matplotlib) - required
2. Installs xvfb virtual display - optional but recommended
3. Downloads and installs ODA File Converter from opendesign.com - required for DWG support
4. Downloads and installs QCAD dwg2bmp from qcad.org - optional, for high-quality screenshots

ODA Download: https://www.opendesign.com/guestfiles/odafileconverter (free registration required)
QCAD Download: https://qcad.org/en/download (Professional Trial)

Check Environment Status

CODEBLOCK3

Outputs JSON-formatted dependency status, including whether each component is installed, its path, and repair commands if missing.

When to use: Run check-env to diagnose issues when any command fails.

Core Tool

All operations use a single CLI tool located at:

CODEBLOCK4

Usage pattern:

CODEBLOCK5

All commands output structured JSON to stdout, making the results directly parseable.

Available Commands

1. info — Get Drawing Basic Information

Read file version, units, entity count summary, and metadata.

CODEBLOCK6

Output includes: DXF version, drawing units, layer count, block count, entity type breakdown.

When to use: As the first step when analyzing any new DWG file. Always start with info to understand the drawing's scope.

2. layers — List All Layers

List all layer definitions with color, linetype, and state (on/off/locked/frozen).

CODEBLOCK7

Options:

• - --count-entities: Count entities on each layer (slower but informative)
• INLINECODE19: Sort layers by name or color

When to use: To understand the drawing's organizational structure and identify which layers contain relevant content.

3. entities — List Model Space Entities

List entities with optional type and layer filtering.

CODEBLOCK8

Options:

• - --type: Filter by entity type (INSERT, LINE, CIRCLE, TEXT, MTEXT, LWPOLYLINE, ARC, etc.)
• INLINECODE22: Filter by layer name (exact match)
• INLINECODE23: Maximum number of results

When to use: To inspect specific types of entities or explore a particular layer's contents.

4. blocks — List Block Definitions

List all block definitions (templates) with their internal entity composition.

CODEBLOCK9

Options:

• - --name-filter: Filter blocks by name (case-insensitive substring match)

When to use: To catalog available equipment types and standard components in the drawing.

5. inserts — List Block Reference Instances

List block reference instances (placed equipment/components) in model space.

CODEBLOCK10

Options:

• - --name-filter: Filter by block name
• INLINECODE28: Show internal layer distribution of each block
• INLINECODE29: Maximum results

When to use: To find specific equipment instances, their positions, and verify their layer assignments.

6. texts — Extract Text Content

Extract all text content (TEXT, MTEXT, and block attributes).

CODEBLOCK11

Options:

• - --keyword: Filter by keyword (case-insensitive)
• INLINECODE32: Maximum results

When to use: To find labels, annotations, equipment names, or any textual information in the drawing.

7. layer-content — Extract Layer Entity Details

Get all entities on a specific layer with full details.

CODEBLOCK12

When to use: To deeply inspect the contents of a particular professional layer.

8. spaces — View Space Layout System

List all spaces (model space, paper space, block definitions).

CODEBLOCK13

Options:

• - --detail: Include viewport information for paper spaces

When to use: To understand the drawing's overall structure and printing layout.

9. distance — Calculate Distance Between Points/Entities

Calculate 2D and 3D distances between two entities or coordinates.

CODEBLOCK14

Output includes: 2D distance, 3D distance, delta X/Y/Z, and unit.

When to use: To verify clearance requirements, check equipment spacing, or measure distances between specific points.

10. screenshot — Entity/Region Screenshot

Capture a visual image of a specific entity, layer, or region.

CODEBLOCK15

Options:

• - --block-name: Center on a named block instance
• INLINECODE39: Center on entity by handle
• INLINECODE40: Capture the bounding box of all entities on a layer
• INLINECODE41: Specify exact capture region
• INLINECODE42: Expansion radius around entity center (default: 5000)
• INLINECODE43: Output file path (supports .png, .pdf, .svg)
• INLINECODE44: Image resolution (default: 3000)
• INLINECODE45: Background color (default: black)
• INLINECODE46: Path to QCAD dwg2bmp tool

When to use: To visually inspect equipment, verify spatial arrangements, or generate images for reports and multimodal AI analysis.

11. audit — Drawing Compliance Audit

Run automated compliance checks on the drawing.

CODEBLOCK16

Checks performed:

• - Zero-layer violations: INSERT entities on default layer 0
• Block layer mixing: Blocks with non-standard internal layer assignments
• Empty layers: Layers with no entities (potentially redundant)
• DEFPOINTS entities: Non-printing entities on DEFPOINTS layer

Output: Issue list with severity (error/warning/info), rule name, and details.

When to use: Before submitting drawings for review, or when auditing received drawings for compliance.

12. search — Search Entities by Keyword

Search entities by keyword across block names, text content, and layer names.

CODEBLOCK17

When to use: To quickly locate specific equipment, annotations, or components by name.

13. export-pdf — Export to PDF

Convert the drawing to a vector PDF file.

CODEBLOCK18

When to use: To generate a portable visual representation of the drawing.

Recommended Workflow

When analyzing a new DWG file, follow this order:

1. 0. Review past learnings → Check .learning/ for relevant experience (see Self-Learning Mechanism)
2. info → Understand file basics (version, units, entity count)
3. layers → See the organizational structure
4. inserts or entities → Find specific equipment/components
5. texts → Extract textual annotations
6. distance → Measure spatial relationships
7. screenshot → Capture visuals for detailed inspection
8. audit → Check compliance

CAD Domain Knowledge

For detailed understanding of DWG file structure, entity types, layer conventions, and CAD coordinate systems, refer to:

CODEBLOCK19

This reference covers: DWG file hierarchy, entity type catalog, unit systems, layer naming conventions for data center drawings, and common audit rules.

Output Format

All commands produce JSON output to stdout. Error messages go to stderr. Exit code 0 indicates success, non-zero indicates failure.

Example output structure (for info command):

CODEBLOCK20

Error Handling

• - If a file does not exist, the tool exits with a descriptive error message
• If required Python packages are missing, the tool outputs clear installation instructions and exits
• If ODA/QCAD are not available, the tool provides manual setup instructions
• If QCAD is not available for screenshots, the tool automatically falls back to matplotlib rendering
• All errors are written to stderr in JSON format for programmatic consumption
• No packages are installed automatically — all installation requires explicit user action

Notes

• - The {SKILL_DIR} placeholder refers to the directory containing this SKILL.md file
• First run may take 1-3 minutes as the tool automatically installs all dependencies
• After first setup, a marker file is created at assets/.setup_done to skip future setup
• To re-run setup (e.g. after system update), delete assets/.setup_done and run any command
• DWG files require ODA File Converter for reading; DXF files can be read directly
• Screenshot quality is best with QCAD dwg2bmp; matplotlib is a reasonable fallback
• Large drawings may take several seconds to load — this is normal for complex engineering files
• Coordinate values in CAD drawings can be very large (millions); this is normal for real-world projects

Self-Learning Mechanism

This skill supports a self-learning mechanism that records user preferences, error resolutions, and best practices discovered during usage. These records are stored in a .learning/ directory within the user's project and are automatically referenced in subsequent sessions to provide more accurate and personalized assistance.

⚡ Core Principle

Unless the user explicitly requests to skip past experience (e.g., "don't refer to previous learnings", "start fresh"), the agent MUST review .learning/ files at the beginning of every task and apply relevant knowledge throughout the session.

📏 Capacity Limits & Eviction

To prevent records from growing indefinitely and losing their effectiveness, strict limits are enforced:

File	Max Entries	Eviction Trigger
INLINECODE65	50	When adding a new entry would exceed 50
INLINECODE66

30 | When adding a new entry would exceed 30 |

Counting method: Count the number of ## [ heading lines in the file to determine the current entry count.

Eviction strategy — when the limit is reached, before adding a new entry, remove the lowest-value entry using this priority order:

1. 1. Status = resolved entries are evicted first (already addressed, least future value)
2. Status = wont_fix entries are evicted next
3. Among candidates with the same status, evict the one with the oldest Logged date
4. Among candidates with the same status and date, evict the one with the lowest priority (low < medium < high < critical)

Important rules:

• - pending and in_progress entries are never evicted automatically — if all entries are pending/in_progress and the limit is reached, the agent should ask the user which entry to remove or resolve
• When evicting, remove the entire entry block (from ## [ to the next ---)
• User preference entries (user_preference category) have eviction protection — they are only evicted when no other candidates remain, because preferences are persistently valuable

Quick count check (the agent should do this before adding a new entry):
CODEBLOCK21

First-Use Initialization

Before logging anything, ensure the .learning/ directory and files exist in the user's project root (NOT in the skill directory). If any are missing, create them:

CODEBLOCK22

Never overwrite existing files. This is a no-op if .learning/ is already initialized.

When to Review Learnings (Start of Session)

At the start of every CAD analysis task, before executing any commands:

1. 1. Check if .learning/ directory exists in the project root
2. If it exists, read .learning/LEARNINGS.md and INLINECODE84
3. Identify entries relevant to the current task (by file type, command, layer names, error patterns, etc.)
4. Apply relevant learnings proactively — for example:

- If a user preference for screenshot background color was recorded, use it automatically - If a known error with certain DWG versions was logged, warn the user or apply the fix preemptively - If a best practice for specific layer analysis was discovered, follow it

When to Log

Situation	Action
Command fails or produces unexpected output	Log to INLINECODE85
User corrects the agent's approach

Log to .learning/LEARNINGS.md with category correction | | User specifies a preference (e.g., output format, default options) | Log to .learning/LEARNINGS.md with category user_preference | | A non-obvious solution is discovered | Log to .learning/LEARNINGS.md with category insight | | Agent's knowledge about CAD was incorrect or outdated | Log to .learning/LEARNINGS.md with category knowledge_gap | | A better workflow or approach is found | Log to .learning/LEARNINGS.md with category best_practice |

Learning Entry Format

Append to .learning/LEARNINGS.md:

CODEBLOCK23

Error Entry Format

Append to .learning/ERRORS.md:

CODEBLOCK24
Actual error message or output
CODEBLOCK25

ID Generation

Format: TYPE-YYYYMMDD-XXX

• - TYPE: LRN (learning), ERR (error)
• YYYYMMDD: Current date
• XXX: Sequential number (e.g., 001, 002)

Examples: LRN-20260408-001, INLINECODE104

Resolving Entries

When an issue is fixed or a learning is confirmed, update the entry:

1. 1. Change **Status**: pending → INLINECODE106
2. For errors, fill in the ### Resolution section

Privacy & Safety

• - Do NOT log secrets, tokens, private keys, or sensitive environment variables
• Prefer short summaries over full command output
• Do NOT log full file contents — reference file paths instead
• The .learning/ directory is local to the project and is NOT committed to version control by default

Gitignore Recommendation

Add to the project's .gitignore:

CODEBLOCK26

Quick Reference for Agent Behavior

CODEBLOCK27