Skip to content

saeedahmadicp/ti-edgeai-skills

v0.1.0

Skills for developing, deploying and profiling neural networks on Texas Instruments Edge AI processors (C7x/MMA with TIDL: TDA4VM, AM68A, AM69A, AM67A, AM62A) using edgeai-tidl-tools and edgeai-gst-apps

Changelog

Format: Keep a Changelog. Versions apply to the plugin as a whole; each skill also carries its own metadata.version.

[0.1.0] - 2026-10-02

First release.

Added

  • Five skills for the TI Edge AI SDK (C7x/MMA with TIDL): ti-edgeai-dev, ti-edgeai-import-model, ti-edgeai-generate-config, ti-edgeai-profile-pipeline, ti-edgeai-train-model.
  • SoC lookup table (TDA4VM, AM68A, AM69A, AM67A, AM62A) and SoC parameters in the scripts (TOOLS_SOC, TIDL_TAG, GST_SOC, --platform, --quant-scale-type, --target-device).
  • Tools: ONNX preflight (advisory), TIDL compile in Docker, detection/classification evaluators, artifact version check, packaging, non-destructive deployment with a meaningful exit status, host-vs-board output comparison, config generator and validator, camera discovery, model/layer/pipeline profiling, a model manifest with file hashes, staged compile output (a failed compile keeps the previous artifacts).
  • Claude Code and Codex installation (scripts/install.sh --agent claude|codex --scope user|repo), optional agents/openai.yaml metadata per skill, root plugin.json (portable layout).
  • Repository lint (scripts/validate_skills.py), unit tests (tests/run_tests.sh, plus tests/ in each skill), eval prompts, docs/ (architecture, testing, verification status), CI workflow, plugin manifest, contribution guide, issue/PR templates.

Verified

  • On an SK-TDA4VM with Edge AI SDK 11.0: tools container build, detector and classifier compile + deploy, host-emulation vs board outputs, image/video/classification/mosaic configs, profiling tools. Everything else is documentation-derived and labelled "not verified" (docs/verification-status.md).