fab-plugins

Portable coding-loop & working-method skills: loopit, deliberate, prioritize, ship, plan-and-decompose, triz, creative-thinking, visual-plan/recap, catch-up, handover, and Svelte/Rust working aids.

Host-bound A2A mesh + self-hosted-service skills: fab-agent-runtime peer lifecycle, mesh-context, graphfusion, gitea/gitea-pm/gitea-bots, woodpecker, bifrost, tensorzero-gateway, openobserve. Opt-in (host-specific).

Quant value-stream skills: fab-swarm-trading, indicator-creator, quant-consult, efficient-frontier, fable-efficient, cosmos-gl, casbin-ecosystem, pi-coding-agent, moshi-best-practices.

Strix-box-only reference skills: llama.cpp (ROCm/Vulkan), vLLM (historical), ROCm profiling, RDNA 3.5 / Qwen 3.6 architecture, hipBLAS internals, Triton kernels, PyTorch-ROCm, container/toolbox ML stacks. Opt-in (machine-specific).

Portable ML eval/bench/training skills: model-guide/picker/quantization, code-bench, niah-bench, llm-eval-overview, thinking-eval, huggingface-workflow, gpu-bench-pipeline, mlflow-experiments, model-training, gepa, skillopt-rust-bugfix.

Cross-repo-duplicated Claude-Flow community skills reconciled to one canonical (newest) version: github-* (code-review, multi-repo, project/release-management, workflow-automation), swarm-* (advanced, orchestration), skill-builder/creator. Sourced from the neural-trader checkout via build.py. Opt-in.

Diagnostics + autonomy for the A2A mesh / local-inference stack: mesh-doctor & recon agents, /mesh-health /system-recon /strix-facts /autonomy skills, bundled mesh-health-probe. One-call diagnosis (vs a fifteen-turn investigation) + decide-and-proceed rubric. Opt-in (host-specific).