Skill Package Specification

The skill definition format follows the open standard of Agent Skills. A skill is an instruction-based asset: a Markdown guide that an LLM reads and follows. Tool (function calling) definition, registration, and execution are handled by the application layer and are invisible to DolphinX.

1. Skill Package Structure

A skill package is a ZIP file, up to 30 MB after extraction, and must include a SKILL.md file located either at the root or in the sole top-level subdirectory:

my-skill/
 SKILL.md # Required: YAML frontmatter + Markdown instruction body
 template.md # Optional: template files
 reference.md # Optional: reference documentation
 scripts/ # Optional: scripts
 helper.py
 assets/ # Optional: resource files

2. SKILL.md Format

It consists of two parts: YAML frontmatter, delimited by ---, and a Markdown body:

---
name: search-database
version: 1.0.0
description: Search distributed tables
argument-hint: "SELECT fields FROM table WHERE condition"
category: database
displayName: Database Search
license: MIT
compatibility: ">=3.0.0"
disable-model-invocation: false
user-invocable: true
---
## Usage Guide
You can use this skill to query DolphinDB distributed tables.
### Parameters
- `$ARGUMENTS`: The SQL query to execute
### Context Variables
- `${SESSION_ID}`: Current session ID
- `${AGENT_ID}`: Current agent ID
- `${USER_ID}`: Current DolphinDB user ID

3. Frontmatter Field Description

Standard fields, as defined by the open standard of Agent Skills:


Field	Type	Required	Description
name	string	Yes	The skill identifier. It must be globally unique. Format: `^[a-z0-9]([a-z0-9-]*[a-z0-9])?$`, ≤64 characters.
description	string	Yes	A skill description that explains what the skill does and when to use it (≤1,024 characters). Injected into the skill directory so the LLM can determine whether the skill is needed.
version	string	No	The skill version, such as 1.0.0.
license	string	No	The license information.
compatibility	string	No	The environment requirements (≤500 characters).
metadata	map	No	User-defined key-value pairs.

DolphinX extension fields, parsed from frontmatter and stored as separate columns:


Field	Type	Default Value	Description
argument-hint	string	—	The parameter hints, such as `[query] [limit]`, displayed in the skill directory.
disable-model-invocation	bool	false	When set to true, the LLM cannot auto-discover the skill, so it is excluded from the skill directory. Users or applications can still trigger it explicitly.
user-invocable	bool	true	When set to true, users can invoke the skill directly with a command.
category	string	—	The skill category tag.
displayName	string	—	The skill name.

All other frontmatter fields, such as model, effort, and allowed-tools, are stored together in metadata JSON. DolphinX does not create separate columns for them.

4. Progressive Disclosure Model


Loading Stage	When to Load	Content	Token Cost
Skill Directory	Injected into the context by default	name+description+argument-hint	Approximately 50–100 per skill
Detailed Skill Instructions	When explicitly loaded by the application layer, for example, through activeSkillNames or user-defined tools or logic	The complete Markdown body of SKILL.md	Recommended: <5,000
Skill Files	When the application layer retrieves them on demand, using primitives such as skill.files.get or skill.files.batchGet	Scripts, templates, reference documentation, and similar files in the package	Unlimited

5. Built-In DolphinX Skills

The following skills are built into DolphinDB Server 3.00.6. After initializing the AI resources on the web interface, you can enable the required skills without any manual installation. For detailed information about each skill, see DolphinX_Skill.


Skill or Skill Package Name	Description
dolphindb-backtest	Used to generate backtesting scripts based on the backtesting plugin, enabling DolphinX to generate DolphinDB backtesting scripts directly from quantitative strategy descriptions.
dolphindb-coding-style	Used to guide DolphinDB code development. It includes the following skills: dolphindb-basic-syntax-programming: Provides a ruleset for basic DolphinDB syntax, script scaffolding, and initial code generation. dolphindb-vectorization-programming: Guides vectorized programming in DolphinDB. dolphindb-functional-programming: Guides functional programming in DolphinDB. dolphindb-sql-programming: Guides writing DolphinDB SQL statements.
dolphindb-data-import	Guides importing data into DolphinDB, covering scenarios such as importing CSV/TXT text files, binary files, data via the HDF5 plugin, data via the Parquet plugin, and data from the IOTDB engine.
dolphindb-factor	Used for factor generation and includes the following skills: dolphindb_daily_factor: Generates factors from daily data. dolphindb_highfreq_factor: Generates factors from intraday/high-frequency data. dolphindb_rag: Enforces source tracing for DolphinDB-related questions.
dolphindb-ficc-pricing	Used to construct instruments for 13 asset classes, including bonds, interest rates, foreign exchange, commodity futures options, and equity options; to build interest rate curves, volatility surfaces, and dividend curves; and to describe the invocation specifications for dedicated pricers and `instrumentPricer`.
dolphindb-hfdataimport	Used to import high-frequency historical market data and includes the following skills: dolphindb-csmar-import: Imports CSMAR historical market data. dolphindb-datayes-import: Imports DataYes historical market data.
dolphindb-machinelearning	Used for data source discovery, raw modeling data selection, field role inference, data cleansing, feature engineering, classification, regression, and clustering.
dolphindb-ops	Used for DolphinDB O&M scenarios, including fault diagnosis for issues such as OOM, slow queries, stream latency, full disks, replication exceptions, and metadata corruption, as well as backup and recovery, security configuration, and job management.
dolphindb-orca-dstream	Guides writing, reviewing, and fixing DolphinDB Orca declarative DStream API code.
dolphindb-plot	Guides generating visualizations such as line charts, column charts, bar charts, pie charts, area charts, scatter plots, and 3D surface plots.
dolphindb-report-factor-replication	Used to reproduce research report factors, covering candidate factor extraction, single-factor logic extraction, field auditing and data retrieval, MR/Execute mode determination, DolphinDB factor code generation, reuse of built-in indicators, common reproduction patterns, and checks for future data leakage.
dolphindb-test	Guides generating DolphinDB functional test plans, test scenario documents, and complete test cases.
dolphindb-tushare-import	Imports Tushare data into DolphinDB.
dolphindb-universal-financial-data	Used to process financial data and includes the following skills: dolphindb-finance-dbtbcreate: Creates databases and tables for financial data in DolphinDB. dolphindb-finance-dataimport: Imports CSV files into existing DolphinDB tables.