seekdb Documentation

Provides comprehensive access to seekdb database documentation through a centralized catalog system.

Quick Start

1. Locate skill directory (see Path Resolution below)

2. Load full catalog (1015 documentation entries)

3. Match query to catalog entries semantically

4. Read document from matched entry

Path Resolution (Critical First Step)

Problem: Relative paths like ./seekdb-docs/ are resolved from the current working directory, not from SKILL.md's location. This breaks when the agent's working directory differs from the skill directory.

Solution: Dynamically locate the skill directory before accessing docs.

Step-by-Step Resolution

1. Read SKILL.md itself to get its absolute path: read(SKILL.md) // or any known file in this skill directory

2. Extract the directory from the returned path: If read returns: /root/test-claudecode-url/.cursor/skills/seekdb/SKILL.md Skill directory: /root/test-claudecode-url/.cursor/skills/seekdb/

3. Construct paths using this directory: Catalog path: <skill directory>references/seekdb-docs-catalog.jsonl Docs base: <skill directory>seekdb-docs/

Documentation Sources

Full Catalog

• Local: <skill directory>references/seekdb-docs-catalog.jsonl (1015 entries, JSONL format)

• Remote: https://raw.githubusercontent.com/oceanbase/seekdb-ecology-plugins/main/agent-skills/skills/seekdb/references/seekdb-docs-catalog.jsonl (fallback)

• Entries: 1015 documentation files

• Coverage: Complete seekdb documentation

• Format: JSONL - one JSON object per line with path and description

Complete Documentation (Local-First with Remote Fallback)

Local Documentation (if available):

• Base Path: <skill directory>seekdb-docs/

• Size: 7.4M, 952 markdown files

• Document Path: Base Path + File Path

Remote Documentation (fallback):

• Base URL: https://raw.githubusercontent.com/oceanbase/seekdb-doc/V1.1.0/en-US/

• Document URL: Base URL + File Path

Strategy:

1. Locate: Determine <skill directory> using path resolution above

2. Load: Load full catalog (1015 entries) - try local first, fallback to remote

3. Search: Semantic search through all catalog entries

4. Read: Try local docs first, fallback to remote URL if missing

Workflow

Step 0: Resolve Path (Do this first!)

# Read this file to discover its absolute path
read("SKILL.md")

# Extract directory from the path

# Example: /root/.claude/skills/seekdb/SKILL.md → /root/.claude/skills/seekdb/

Step 1: Search Catalog

Start with grep for keyword searches. Only load full catalog when necessary.

Method 1: Grep Search (Preferred for 90% of queries)

Use grep to search for keywords in the catalog:

grep -i "keyword" <skill directory>references/seekdb-docs-catalog.jsonl

Examples:

# Find macOS deployment docs
grep -i "mac" references/seekdb-docs-catalog.jsonl

# Find Docker deployment docs
grep -i "docker\|container" references/seekdb-docs-catalog.jsonl

# Find vector search docs
grep -i "vector" references/seekdb-docs-catalog.jsonl

Method 2: Load Full Catalog (Only when necessary)

Load the complete catalog only when:

• Grep returns no results

• Complex semantic matching is required

• No specific keyword to search

Local: <skill directory>references/seekdb-docs-catalog.jsonl
Remote: https://raw.githubusercontent.com/oceanbase/seekdb-ecology-plugins/main/agent-skills/skills/seekdb/references/seekdb-docs-catalog.jsonl (fallback)
Format: JSONL (one JSON object per line)
Entries: 1015 documentation files

Strategy:

1. Try local catalog first: <skill directory>references/seekdb-docs-catalog.jsonl

2. If local missing, fetch from remote URL above

Catalog contents:

• Each line: {"path": "...", "description": "..."}

• All seekdb documentation indexed

• Optimized for semantic search and grep queries

Step 2: Match Query

Analyze search results to identify the most relevant documents:

For grep results:

• Review matched lines from grep output

• Extract path and description from each match

• Select documents whose descriptions best match the query

• Consider multiple matches for comprehensive answers

For full catalog:

• Parse each line as JSON to extract path and description

• Perform semantic matching on description text

• Match by meaning, not just keywords

• Return all relevant entries for comprehensive answers

Note: The catalog contains path and description fields. The description field contains topic and feature keywords, making it suitable for both keyword and semantic matching.

Step 3: Read Document

Local-First Strategy:

1. Try local first: <skill directory>seekdb-docs/[File Path]

   • If file exists → read locally (fast)
   • If file missing → proceed to step 2

2. Fallback to remote: https://raw.githubusercontent.com/oceanbase/seekdb-doc/V1.1.0/en-US/[File Path]

   • Download from GitHub

Example:

Query: "How to integrate with Claude Code?"

1. Resolve path: read(SKILL.md) → /root/.claude/skills/seekdb/SKILL.md
   Skill directory : /root/.claude/skills/seekdb/

2. Search catalog with grep:
   grep -i "claude code" /root/.claude/skills/seekdb/references/seekdb-docs-catalog.jsonl

3. Match query from grep results:
   → Found: {"path": "300.integrations/300.developer-tools/700.claude-code.md",
             "description": "This guide explains how to use the seekdb plugin with Claude Code..."}
   → This matches the query, select this document

4. Read doc:
   Try: /root/.claude/skills/seekdb/seekdb-docs/300.integrations/300.developer-tools/700.claude-code.md
   If missing: https://raw.githubusercontent.com/oceanbase/seekdb-doc/V1.1.0/en-US/300.integrations/300.developer-tools/700.claude-code.md

Guidelines

• Always resolve path first: Use the read-your-SKILL.md trick to get the absolute path

• Prefer grep for keyword queries: Load full catalog only when grep returns nothing or semantic matching is needed

• Semantic matching: Match by meaning, not just keywords

• Multiple matches: Read all relevant entries for comprehensive answers

• Local-first with remote fallback: Try local docs first, use remote URL if missing

• Optional local docs: Run scripts/update_docs.sh to download full docs locally (faster)

• Offline capable: With local docs present, works completely offline

Catalog Search Format

The catalog file is in JSONL format (one JSON object per line):

{"path": "path/to/document.md", "description": "Document description text"}

Searching the catalog:

• Keyword search: Use grep (see Step 1 examples). Each matched line contains both path and description.

• When grep is insufficient: Read the full catalog, parse each line as JSON, then do semantic matching on descriptions.

Common Installation Paths

This skill may be installed at:

• Cursor: .cursor/skills/seekdb/

• Claude Code: .claude/skills/seekdb/

• Custom: Any directory (path resolution handles this automatically)

Do not hardcode these paths. Use the dynamic resolution method instead.

Detailed Examples

See examples.md for complete workflow examples including:

• Full catalog search scenarios

• Local-first lookup scenarios

• Remote fallback scenarios

• Integration queries

• Multi-turn conversations

Category Overview

• Get Started: Quick start, basic operations, overview

• Development: Vector search, hybrid search, AI functions, MCP, multi-model

• Integrations: Frameworks, model platforms, developer tools, workflows

• Guides: Deployment, management, security, OBShell, performance

• Reference: SQL syntax, PL, error codes, SDK APIs

• Tutorials: Step-by-step scenarios