OpenParable
Skip to content
← Back to skills

Review Spell

Testing & Quality
review-spell

Review Spell is an Agent Skill for testing & quality published in the duneanalytics/spellbook repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.

Review Spellbook spell contributions for config blocks, schema YAML, SQL style, Jinja, performance, seeds, architecture, and compile/test readiness.

View on GitHub ↗
QUICK START

How to use this skill

Bring this guide into your coding agent with a prompt tailored to the tool you use.

  1. Open your project in Codex.
  2. Copy and paste the prompt below.
  3. Review files and risks before approving any action.
Prompt to paste
Inspect the Agent Skill at https://github.com/duneanalytics/spellbook/blob/HEAD/.claude/skills/review-spell/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/review-spell, but wait for my approval before writing files or running scripts.

Copying this prompt does not install or run the skill. Review third-party files before use.

Review Spell PR

Use this skill when reviewing a Spellbook spell contribution (new model or modification). Walk through each section below as a checklist.

1. Config Block

  • schema is present and correct for the Dune namespace
  • alias is present and matches intended table/view name
  • materialized is explicitly declared (not relying on dbt_project.yml default)
  • If table or incremental: file_format='delta' is present
  • If incremental:
    • incremental_strategy specified (merge, append, or delete+insert)
    • unique_key specified — no columns that could contain NULLs
    • incremental_predicates uses the incremental_predicate() macro (unless full-history lookup needed)
    • If partitioned: partition column(s) included in unique_key
  • Config block formatting follows repo SQL conventions (tabs, trailing comma on last param, {{ config( on first line)

2. Schema YML (_schema.yml)

  • Model entry exists with name matching the SQL filename
  • Model has a description
  • dbt_utils.unique_combination_of_columns test present — columns match config unique_key exactly
  • not_null test on each unique key column
  • Key columns have descriptions
  • For sector-level spells: seed test is present (e.g., check_dex_base_trades_seed)

3. SQL Style

  • Leading commas (left comma club)
  • Tab indentation (no spaces)
  • All SQL keywords and function names lowercase
  • New line after select, from, where, group by, order by, etc.
  • Explicit join types (inner join, left join — never bare join)
  • Table aliases use as keyword (from users as u, not from users u)
  • All columns prefixed with table aliases when joins are present
  • CTEs use leading commas between them, with and CTE name on same line

4. Jinja

  • All table references use source() or ref() — no hardcoded table names
  • Jinja whitespace: trailing - only ({% if -%}, {% else -%}, {% endif -%})
  • {% if is_incremental() -%} block present for incremental models with:
    • Incremental path using {{ incremental_predicate('source.block_time') }}
    • Non-incremental path with earliest date filter

5. Performance

  • Join order: larger table on left side
  • Partition filters present in WHERE clauses (block_date, block_time)
  • Cross-chain tables filtered by both blockchain and time
  • No SELECT * on large tables
  • UNION ALL used (not bare UNION) unless deduplication truly needed
  • No ORDER BY without LIMIT on large result sets
  • Time filters in both ON and WHERE clauses when joining on partition columns

6. Seed File (Sector-Level Spells)

  • Seed CSV present with representative rows
  • Seed registered in directory's _schema.yml with column types
  • Seed columns include all unique key columns + fields to test
  • Seed is small (handful of rows, not hundreds)
  • Seed test called in model's _schema.yml with correct parameters

7. Architecture (Sector-Level Spells)

  • Follows lineage: platform base → chain-level union (table) → cross-chain union (view) → final enriched spell
  • Platform base spells use macros for forked protocols where applicable
  • One model per protocol, per version, per blockchain
  • Metadata enrichment saved for downstream (base spells contain raw data only)

8. Compile & Test

Run dbt compile in the relevant sub-project to verify:

  • No compilation errors
  • Compiled SQL in target/ looks correct
  • Test with python scripts/dune_query.py "@model_name" --limit 100 or paste into Dune