AI coding agent skills for YDB — for writing YQL, designing schemas, and reviewing Java application code against YDB. Skills auto-trigger based on context.
Status: framework and scaffolding. Skill bodies are intentionally thin — content is migrated surface-by-surface, grounded in live YDB SDK source and upstream docs.
| Skill | Scope |
|---|---|
| ydb-core | Entry point / router. YDB overview, auth and connection, schema basics, admin CLI. Baseline skill — auto-installed alongside any other. |
| ydb-table | Writing YQL and executing it (SDK-embedded, ydb sql / ydb yql CLI). Optimization, schema design for query patterns, SQL-to-YQL conversion, audit. |
git clone https://github.com/ydb-platform/ydb-ai-skills.git
cd ydb-ai-skills
# Auto-detect agents in current project
./install.sh --detect
# Install for specific agent
./install.sh --agent=claude
# Install only ydb-table (ydb-core auto-included as baseline)
./install.sh --agent=claude --skills=ydb-table
# Install only ydb-table without the baseline
./install.sh --agent=claude --skills=ydb-table --no-core
# Dry run — see what would be done
./install.sh --agent=claude --dry-run# Auto-detect agents
curl -fsSL https://ai.ydb.sh | bash
# Specific agent
curl -fsSL https://ai.ydb.sh | bash -s -- --agent=cursor
# Multiple agents
curl -fsSL https://ai.ydb.sh | bash -s -- --agent=claude,copilot,gemini
# Global (user-level)
curl -fsSL https://ai.ydb.sh | bash -s -- --agent=claude --global
# All options
curl -fsSL https://ai.ydb.sh | bash -s -- --help| Agent | Project dir | Global dir |
|---|---|---|
| Claude Code | .claude/skills/ |
~/.claude/skills/ |
| Cursor | .cursor/skills/ |
— |
| Windsurf | .windsurf/skills/ |
— |
| GitHub Copilot | .github/skills/ |
~/.copilot/skills/ |
| Codex CLI | .agents/skills/ |
~/.codex/skills/ |
| Roo Code | .roo/skills/ |
— |
| Gemini CLI | .gemini/skills/ |
~/.gemini/skills/ |
| Amp | .agents/skills/ |
~/.config/agents/skills/ |
| Kiro | .kiro/skills/ |
— |
| Trae | .trae/skills/ |
— |
| Generic | .agents/skills/ |
~/.agents/skills/ |
Skills trigger automatically from the user's phrasing. Examples of queries that route to each skill:
> What is YDB and how do I connect to it from Go?
→ ydb-core
> Write a YQL query to paginate users by created_at
→ ydb-table
> Review this Java/Hibernate code that calls saveAll in a loop
→ ydb-table (audit mode)
For agents that don't auto-trigger skills, reference the skill name explicitly in the prompt.
skills/ Surface-aligned skills (universal format)
ydb-core/SKILL.md Single-file router — overview, auth, schema basics, admin CLI
ydb-table/ SKILL.md + references/ + rules/
promptfooconfig.yaml Compatibility matrix config — provider list, test discovery
prompts/coding-agent.yaml Shared system prompt for all tests
tests/ Test cases per skill (YAML, one per file)
agents/grader.md Reference grader prompt (from anthropics/skills)
docs/
authoring.md How to write a skill — structure, conventions, review checklist
testing.md How to run the compatibility matrix with promptfoo
templates/ Copy-paste skeletons for SKILL.md, references, rules
install.sh Universal installer
A compatibility matrix built with promptfoo. Each (provider, model) pair runs every test; the grader scores each result via llm-rubric. Identifies the minimum model per vendor family that still works.
Quick run:
export OPENROUTER_API_BASE_URL="..."
export OPENROUTER_API_KEY="..."
npx promptfoo@latest eval
npx promptfoo@latest viewAll providers route through one OpenAI-compatible endpoint. Default model set skews toward enterprise-deployed open-source (Qwen, DeepSeek, Mistral, Llama, GLM, Kimi) with one baseline per major cloud vendor (Anthropic, Google, OpenAI, xAI).
Details — adding tests, adding models, reading the matrix, known gaps — in docs/testing.md.
Read docs/authoring.md before adding content. The short version: stay grounded in upstream YDB SDK source, don't invent rules from memory, one RULE-<PREFIX>-<NN> prefix per category claimed in the registry on first use, keep SKILL.md bodies short.
Apache-2.0