Add AQUA skills directory for coding agents

[Aryanag2]

Motivation

Coding agents are increasingly being used to help developers write, debug, and operate software — and ML engineers are no exception. Developers building on OCI AI Quick Actions (AQUA) are already using coding agents to help them register models, configure deployments, write fine-tuning jobs, and debug errors. However, without structured context about AQUA's APIs, CLI, and workflows, agents frequently hallucinate parameter names, miss required fields, or suggest patterns that don't match the actual SDK.

A skills/ directory solves this by giving coding agents a structured, authoritative source of truth about AQUA that they can load on demand — reducing hallucination, cutting back-and-forth, and letting developers stay in flow instead of context-switching to docs.

Summary

Adds a skills/ directory at the repo root containing 8 skills that cover the full AQUA (AI Quick Actions) LLM lifecycle. Skills follow progressive disclosure: lean SKILL.md core files with heavy content offloaded to references/ and examples/ subdirectories. All reference files are explicitly linked from their SKILL.md.

Skills Added

Skill	Description
aqua-model-lifecycle	Register, list, get models from HuggingFace, Object Storage, GGUF, or BYOC
aqua-deployment	Single/multi/stacked deployments, vLLM config, tool calling, shape recommender, LMCache, private endpoints, batch inference
aqua-finetuning	LoRA fine-tuning, dataset formats, hyperparams, distributed training
aqua-evaluation	BERTScore, ROUGE, perplexity evaluation with official sample datasets
aqua-metrics	Prometheus + Grafana monitoring stack via OCI Container Instance
aqua-troubleshooting	OOM, auth, capacity, container error diagnosis patterns
aqua-cli	Complete ads aqua CLI reference with full parameter tables
oci-data-science	Jobs, Pipelines, Model Catalog, LangChain integration, auth patterns

Reference Files

• aqua-deployment/references/shapes.md — GPU shape table, model fit guide, shape recommender CLI/SDK docs
• aqua-deployment/references/lmcache.md — KV cache persistence (LMCache) for multi-turn workloads
• aqua-deployment/references/private-endpoints.md — Private endpoint creation and usage
• aqua-deployment/references/batch-inferencing.md — Job-based offline batch inference with vLLM
• aqua-model-lifecycle/references/containers.md — Current container versions (vLLM 0.11.0, TGI 3.2.1, Llama-cpp 0.3.7)
• aqua-model-lifecycle/references/tgi-migration.md — Migrating TGI-registered models to vLLM
• aqua-cli/references/params.md — Exhaustive parameter tables for all CLI command groups
• aqua-evaluation/examples/ — Official oracle-samples evaluation datasets (no-sys-message + with-sys-message)
• aqua-finetuning/examples/ — Dataset format examples (instruction, conversational, multimodal)

Content Sources

• ads/aqua/ source code (API signatures, entity fields, CLI entry points)
• oracle-samples/oci-data-science-ai-samples official AI Quick Actions documentation