/00. Claude Plugins

A Claude Code plugin is a folder of skills, agents, hooks, and slash commands that Claude installs per project. 28 plugins, 68 skills here.

/plugin marketplace add aiocean/claude-plugins
/plugin install <plugin-name>@aiocean-plugins

Skills load when their description matches your message, hooks fire on tool-call events, agents spawn on the Agent tool. Idle plugins cost nothing — browse plugins or read the guides.

Folders

Articles

/02. aio-research

Install

/plugin install aio-research@aiocean-plugins

aio-research

Two tools for knowledge work: structured investigation and semantic retrieval.

Research in software projects takes two forms. The first is investigative: you have a question — a technology decision, a market gap, a performance hypothesis — and you need to work through it systematically before committing to an answer. The second is retrieval: you have an existing body of knowledge — documentation, a codebase, a knowledge base — and you need to find the right piece of it quickly, by meaning rather than keyword.

This plugin addresses both. aio-research-kit provides a 10-phase framework for structured investigation. aio-rag-kit connects Claude to a Qdrant vector database for semantic search over any indexed content.

Installation

/plugin marketplace add aiocean/claude-plugins
/plugin install aio-research@aiocean-plugins

Skills

aio-research-kit — Structured 10-phase research workflow

Unstructured research produces unverifiable conclusions. This skill enforces a phase-gated process that separates question formulation from data collection, and data collection from interpretation — the same discipline that makes scientific results reproducible.

"start research", "investigate a topic", "structured research", "deep research", "literature review", "10-phase research", "research workflow"

The 10 phases:

PhaseWhat happens
1. Question FormulationDefine research questions precisely
2. Literature ReviewSurvey existing knowledge and prior work
3. Hypothesis FormationState falsifiable hypotheses
4. Methodology DesignPlan how you will investigate
5. Data CollectionGather evidence systematically
6. Data AnalysisLook for patterns and statistical significance
7. InterpretationDraw conclusions from the analysis
8. ValidationCross-check findings against independent sources
9. DocumentationWrite up findings in full
10. Peer ReviewExternal validation before acting

Initialize a research project with research init, which creates the directory structure and phase templates. Validate completeness with research check.

The CLI tool installs via uv (recommended) or pip from github.com/nguyenvanduocit/research-kit. Optional AI-assisted reasoning during analysis is available via DEEPSEEK_API_KEY (DeepSeek R1) or GOOGLE_AI_API_KEY (Gemini).

Two common patterns where the phase structure prevents common mistakes:

Technical decision research — Phase 1 states the question precisely ("What is the latency impact of switching from REST to gRPC under 1000 RPS?"). Phase 3 forces a falsifiable hypothesis before running benchmarks. Phase 6 analyzes the numbers. Phase 8 cross-checks against existing published benchmarks. The conclusion in Phase 9 is traceable back to evidence, not intuition.

Market research — Phase 2 surveys competitor products before Phase 5 collects feature matrices. This ordering prevents confirmation bias: you survey the landscape first, then gather structured data, then interpret gaps.


aio-rag-kit — Vector search and RAG pipelines over your content

Keyword search fails when the query and the document use different words for the same concept. Semantic search solves this by embedding both query and content into a shared vector space, then finding the nearest neighbors by meaning.

"create vector collection", "index content", "semantic search", "RAG setup", "qdrant", "embeddings", "vector database", "retrieval-augmented generation", "similarity search"

The skill connects to a Qdrant instance (local Docker or Qdrant Cloud) and exposes three operations via MCP:

Collection management — create a collection with the right vector size for your embedding model (1536 for OpenAI text-embedding-3-small), list collections, delete collections.

Indexing — add content with metadata:

rag_index_content(
  collection_name: "docs",
  content: "Document text to index...",
  metadata: {"source": "runbook.md", "section": "deployment"}
)

Search — query by meaning, get ranked results with similarity scores:

rag_search(
  collection_name: "docs",
  query: "How do we roll back a failed deploy?",
  limit: 5
)

A CLI fallback (rag-cli) is available for environments where MCP is not configured. Setup requires Qdrant running (local or cloud) and an OpenAI API key for embeddings, both configured in .mcp.json.

Common use: index a codebase file by file with metadata ({"file": "src/auth.go"}), then query it semantically to find relevant implementation patterns before making a change.

When to use which skill

SituationSkill
Technology decision with significant tradeoffsaio-research-kit
Competitive or market analysisaio-research-kit
Performance hypothesis to validate empiricallyaio-research-kit
"Find where in the codebase X is handled"aio-rag-kit
Build a searchable knowledge base from docsaio-rag-kit
Augment Claude with a private document corpusaio-rag-kit

The two skills compose: use aio-research-kit to structure an investigation, use aio-rag-kit during Phase 5 (Data Collection) to retrieve relevant content from a large corpus that Claude cannot hold in context at once.

Skills (2)

  • aio-rag-kit — Set up vector database collections, index content, and run semantic search via RAG pipelines using Qdrant and rag-kit (auto-installs if missing). Use for retrie…
  • aio-research-kit — Run a structured 10-phase research workflow to investigate a topic, conduct a literature review, or deep-research a technical or market question via research-cl…

Articles