TakeoffLens v0.2.1

What changed

• Updated every canonical repository, release, clone, badge, and Skill URL to anekhirun/Takeoff-Lens-Plugin.
• Added exact Hermes Desktop installation steps for the TakeoffLens Skill and local stdio MCP server.
• Aligned the Plugin manifest, MCP server, candidate-review page, documentation, and tests at v0.2.1.
• Added product architecture, development, installation, changelog, and roadmap documentation from PR #5.

Validation

• 27 unit tests passed.
• MCP doctor passed with TakeoffLens v0.2.1 and 10 tools.
• Plugin and Skill validators passed.
• GitHub Actions passed.
• Release ZIP contains 34 allowlisted entries and no drawings, generated outputs, caches, or machine-local paths.

SHA-256

252DF92ED7B093F61AD356BBEAE2C24425147175150D5FABF30C3121B1D6E607

TakeoffLens v0.2.0

TakeoffLens v0.2.0

TakeoffLens is the new product identity for the reviewed engineering-drawing takeoff Plugin. Electrical and ELV are the active v0.2.0 disciplines, backed by a shared architecture that can add further building-system packs without presenting roadmap scope as implemented.

Highlights

• renamed the Codex Plugin and MCP server to takeoff-lens
• added the active/planned discipline catalog
• added SHA-256 detection provenance and hard failure on mismatched source, template, or candidates
• preserved shortlisted, filtered, and ranked-out candidate pools for audit and regression analysis
• added reviewed precision, recall, F1, and stage-attributed miss evaluation
• separated indoor and weatherproof Fire Alarm variants
• improved layer-aware candidate filtering while retaining mandatory visual review
• added benchmark tooling and public ground-truth examples without including private drawings

Supported scope

• Power
• Lighting
• Fire Alarm
• Data/Voice
• CCTV/Security

Mechanical/HVAC, Plumbing/Sanitary, Fire Protection, Architectural, and Structural discipline packs remain planned.

Validation

• 27 unit tests passed
• MCP doctor passed with 10 tools
• Skill and Plugin validation passed
• packaged archive scan found no PDF, DWG, DXF, output, test-run, cache, or machine-specific artifacts
• installed-package Fire Alarm smoke benchmark: precision 0.60, recall 1.00, F1 0.75, zero false negatives

Installation

Download takeoff-lens-plugin-v0.2.0.zip, extract it, add the included plugin-marketplace directory to Codex, and install takeoff-lens.

When migrating, keep only the takeoff-lens MCP registration and remove legacy drawing-estimate-reader or engineering-drawing-estimator registrations.

SHA-256: 37AB0DC6C04688846686867C220150E222B186BA283B9F6A0EF81A36A505A99A

v0.1.5 — Plugin Distribution

v0.1.5 — Plugin Distribution

• Bundles the reviewed Skill and local MCP server as one versioned Codex Plugin
• Adds relocatable .mcp.json and Windows PowerShell launcher
• Creates a hashed per-user Python runtime and reuses it on later launches
• Separates lightweight headless MCP dependencies from optional PySide6 desktop dependencies
• Includes an allowlisted local-marketplace ZIP for installation and sharing
• Keeps all drawing processing local and preserves mandatory candidate review

Validation:

• 19 regression tests passed locally
• GitHub Actions passed
• source and packaged Plugin validation passed
• clean dependency bootstrap initialized MCP v0.1.5 successfully
• warm plugin startup measured about 0.77 seconds locally
• release ZIP contains 28 entries, one manifest, and no drawings, outputs, virtual environments, caches, or machine-specific paths

Plugin marketplace ZIP SHA256: 1B9C40B8DB4409170536691E3A22BF152FE58BA0728C1835C2275627BD3AB53B

v0.1.4 — Native Pipeline Optimization

v0.1.4 — Native Pipeline Optimization

• Shared native sheet context and vector extraction reuse
• Page Profiler v2 and Candidate Filtering v2
• Deterministic vector-layer signatures with confirmed project mappings
• Compact MCP responses with full audit artifacts retained on disk
• Power, Lighting, Fire Alarm, Data/Voice, and CCTV/Security scope
• Human/agent review remains required before final quantities

Validation: 16 tests passed, MCP doctor passed with 8 tools, and the E3 Lighting benchmark reproduced 10 recessed diffuser and 15 weatherproof IP65 luminaires.

Source ZIP SHA256: 14C3571C3CF0F3260137332F6F95DF5AD8B9DBD1B9EE21FE135206263CA6009C

AI Engineering Drawing Estimator v0.1.3

Highlights

• v0.1.3 scope is focused on Power and Fire Alarm systems.
• Added one-call prepare_sheet_audit workflow for faster per-sheet preparation.
• Added Fire Alarm symbol catalog and project-specific template workflow.
• Added accepted, rejected, uncertain, and unresolved review states.
• Final quantities require review_complete: true; ambiguous points must be shown and confirmed.
• Added an experimental Windows Desktop review interface while MCP + Skill remains the primary workflow.
• Added automated tests and updated installer, documentation, and Skill metadata.

Validation

• 6 unit tests passed
• MCP doctor passed with 7 tools
• Skill validation passed
• Release archive scan: 40 entries, 0 PDF/DWG/DXF or generated-output matches
• SHA-256: 2DDCFE8FA0C861DCE83341061090D199CCB51F68A2DF340308F22BBED78B456F

Important

This remains an assisted-review estimator. Candidate detections are not final quantities until reviewed.

AI Engineering Drawing Estimator v0.1.2

Renames the project to AI Engineering Drawing Estimator and the MCP server to engineering-drawing-estimator.

Includes the Codex/AI Agent skill, local MCP server, installer, candidate review workflow, and symbol-counting tools for engineering drawings.

The release package excludes project drawings, generated outputs, local environments, and private working data.

Drawing Estimate Reader v0.1.1

Initial public release. Includes the AI Agent Skill, local stdio MCP server, redacted starter geometry templates, Windows installer, auditable candidate review, text-overlap suppression, and zero-result review warnings. No project drawings or test outputs are included.