DeepCamera v2026.3 — Skill Platform Milestone
136 commits since the last release, transforming DeepCamera from a monolithic AI camera backend into a decentralized Skill Platform for Aegis-AI.
🏗️ Skill Platform Architecture
- 18 skills across 10 categories: Detection, Analysis, Annotation, Automation, Streaming, Channels, Transformation, Integrations, Camera Providers, and Benchmarking
- Standardized SKILL.md manifest +
requirements.txtdependency model - JSON-lines over stdin/stdout communication protocol
- Progress event emission during startup for Aegis UI feedback
- Graceful SIGTERM shutdown handling
📊 HomeSec-Bench: 131-Test Benchmark Suite
The Home Security AI Benchmark is the anchor skill, evaluating LLM and VLM performance in residential security contexts:
- 131 tests across 16 suites (96 LLM + 35 VLM)
- Context Preprocessing, Knowledge Distillation, Event Deduplication, Security Classification
- Tool Use validation against real Aegis agent-bridge.cjs definitions
- 35 AI-generated security camera fixture frames
- Auto-generated HTML report with Historical Comparison table
- Qwen3.5 compatibility:
<think></think>injection to disable reasoning, balanced-brace JSON parser, buffered token streaming with[C]/[R]field tagging, smart early-abort (100 reasoning tokens, 2× maxTokens, 2000 global cap)
🖥️ Cross-Platform Deployment
Windows
deploy.batequivalents for all skills withdeploy.sh- WMI-based GPU detection fallback
- Quoting bug fixes in GPU detection
AMD ROCm
- ROCm 7.2+ GPU detection
- PyTorch+HIP inference (replacing ONNX)
- Auto-detect installed ROCm version for correct PyTorch wheel
- Graceful CPU fallback when ROCm PyTorch unavailable
Apple Silicon
- YOLO routing to Neural Engine (NPU) via CoreML
- numpy <2.0.0 pin for CoreML export compatibility
🔧 Infrastructure
- CameraClaw registry migrated to DeepCamera with
urlfield - Third-party benchmarks removed from skill registry
- Modernized README with skill catalog and platform identity
src/directory index with module map and architecture docs
📦 Skill Catalog (10 Categories)
| Category | Skills |
|---|---|
| Detection | YOLO 2026 |
| Analysis | Home Security Benchmark, VLM Scene Analysis |
| Annotation | Bounding Box Overlay |
| Automation | Schedule Engine |
| Streaming | go2rtc, RTSP Relay |
| Channels | Discord, Telegram |
| Transformation | Frame Resize, Video Compose |
| Camera Providers | Blink, ONVIF |
| Integrations | CameraClaw (OpenClaw) |
Full Changelog: Aegis...v2026.3