incremental-skills

Teach any AI coding agent how to build and maintain incrementally-refreshed materialized views — engine-agnostic.

Install

npx skills add clickzetta/incremental-skills

What You Get

One master skill (incremental-computation) that covers the core model, source prerequisites, strategy routing, and workflow.

Before You Start

Converting a regular SQL query into incremental computation produces scripts of significantly higher complexity. The more complex the original SQL, the harder the resulting scripts become to read and maintain — this is an inherent property of incremental algorithms, not an implementation shortcoming. If the agent becomes trapped in a prolonged struggle during the conversion process, it may indicate that the agent is not yet fully capable of handling the current transformation task. Forcing the conversion to proceed may introduce hidden defects. We therefore recommend using a capable model when performing the conversion and verifying correctness step by step.

Examples

/incremental-computation Explain incremental computation — how it works, when to use it. Then write a runnable demo.

Assuming the agent knows how to connect to your Spark cluster and read job definitions:

/incremental-computation Convert the batch job order_summary to incremental computation
/incremental-computation REFRESH TABLE order_summary

How It Works

1. Agent loads the incremental-computation skill
2. Confirms how to obtain snapshot, delta, schema, and version for each source table
3. Generates incremental SQL based on the corresponding algorithm references
4. This skill pack focuses on the theoretical principles and algorithms underlying incremental computation across diverse scenarios, rather than providing system-specific implementations. Therefore, before initiating any conversion, ensure the agent has sufficient knowledge of the target big data platform and is capable of performing data development tasks on it. This prevents the agent from becoming trapped in a cycle of iteratively modifying scripts.

Supported Agents

Claude Code, Cursor, Kiro, Copilot CLI, Gemini CLI, Codex — any agent that supports the Agent Skills spec.

Limitations

Since this skill pack only covers the core characteristics of incremental computation, the incremental algorithms for some operators are not the most efficient versions (efficient algorithms require special incremental capabilities developed by big data engines). Additionally, there are scenarios that cannot be supported (e.g., queries with random functions, missing equivalent PK columns, joins with inequality conditions), and the transformed queries may have performance issues (lacking independent performance optimization from incremental engines). If you encounter these situations, especially for complex or business-critical production scenarios with strict requirements for data safety and timeliness, we recommend using a more mature incremental computation product such as Singdata.

License

ClickZetta Incremental Skills License v1.0 — see LICENSE.

Free for non-commercial use. Modifications, derivative works, redistribution, and commercial use are not permitted.