feat(skills): expand writer-sql to codegen-database

.agents/skills/README.md

@@ -10,6 +10,7 @@ A complete skill is a directory with a required `SKILL.md` file and optional bun
 | --- | --- | --- |
 | [`audit-skill-security`](audit-skill-security/SKILL.md) | Auditing third-party or local skills before installing, updating, or trusting them. | [`references/audit-protocol.md`](audit-skill-security/references/audit-protocol.md) |
 | [`codegen-backend`](codegen-backend/SKILL.md) | Production backend code: APIs, services, middleware, workers, persistence, validation, auth, and backend tests. | [`references/`](codegen-backend/references/), [`evals/`](codegen-backend/evals/) |
+| [`codegen-database`](codegen-database/SKILL.md) | Database code: schemas, DDL, OLTP SQL, analytics SQL, migrations, indexes, stored procedures, and dialect-specific scripts. | [`references/`](codegen-database/references/), [`evals/`](codegen-database/evals/) |
 | [`codegen-frontend`](codegen-frontend/SKILL.md) | Production frontend code: components, routes, client state, forms, styling, accessibility, performance, PWA behavior, and visualization. | [`references/`](codegen-frontend/references/), [`evals/`](codegen-frontend/evals/) |
 | [`codegen-test`](codegen-test/SKILL.md) | Automated tests and evals, including E2E, API, integration, performance, AI output, tool-use, RAG, and prompt regression suites. | [`references/`](codegen-test/references/), [`scripts/`](codegen-test/scripts/), [`evals/`](codegen-test/evals/) |
 | [`creator-rule`](creator-rule/SKILL.md) | Writing or improving agent rules, instruction files, `AGENTS.md`, `CLAUDE.md`, Cursor rules, Copilot instructions, and `.agents/rules/*.md`. | [`scripts/`](creator-rule/scripts/), [`evals/`](creator-rule/evals/) |
@@ -21,7 +22,6 @@ A complete skill is a directory with a required `SKILL.md` file and optional bun
 | [`review-code`](review-code/SKILL.md) | Reviewing code changes, diffs, pull requests, branches, or patches for correctness, regressions, security, performance, and test gaps. | [`references/`](review-code/references/), [`evals/`](review-code/evals/) |
 | [`writer-prd`](writer-prd/SKILL.md) | Product requirements, product briefs, feature requirements, product scope, and launch requirements. | [`references/`](writer-prd/references/), [`evals/`](writer-prd/evals/) |
 | [`writer-spec`](writer-spec/SKILL.md) | Technical specs, design docs, functional and non-functional requirements, data contracts, UI specs, release specs, and handoff docs. | [`references/`](writer-spec/references/), [`evals/`](writer-spec/evals/) |
-| [`writer-sql`](writer-sql/SKILL.md) | Database schemas, SQL queries, dialect guidance, normalization, indexing, optimization, and troubleshooting. | [`references/`](writer-sql/references/), [`evals/`](writer-sql/evals/) |
 | [`writer-tech-docs`](writer-tech-docs/SKILL.md) | READMEs, API docs, endpoint references, routine and on-call runbooks, changelogs, and release notes. | [`references/`](writer-tech-docs/references/), [`evals/`](writer-tech-docs/evals/) |
 | [`writer-user-story`](writer-user-story/SKILL.md) | User stories, acceptance criteria, developer tasks, tickets, story points, and sprint planning breakdowns. | [`references/`](writer-user-story/references/), [`evals/`](writer-user-story/evals/) |
 
@@ -41,13 +41,13 @@ Use [`creator-skill`](creator-skill/SKILL.md) to create, revise, package, or eva
 
 ```bash
 cd .agents/skills/creator-skill
-python3 -m scripts.package_skill ../writer-sql /tmp/skills-dist
+python3 -m scripts.package_skill ../codegen-database /tmp/skills-dist
 ```
 
 Use this validation command when changing an existing skill:
 
 ```bash
-python3 .agents/skills/creator-skill/scripts/quick_validate.py .agents/skills/writer-sql
+python3 .agents/skills/creator-skill/scripts/quick_validate.py .agents/skills/codegen-database
 ```
 
 The key rule is simple: keep `SKILL.md` and any files it references together. If a skill says to read `references/postgres.md`, that file must remain available relative to the skill folder. Tiny rule, large consequences. Filesystems enjoy pettiness.

.agents/skills/codegen-database/SKILL.md

@@ -0,0 +1,88 @@
+---
+name: codegen-database
+description: >
+  Generate or modify database code: schemas, DDL, SQL queries, migrations, analytics SQL,
+  indexes, stored procedures, and dialect-specific database scripts.
+author: Oleg Shulyakov
+license: MIT
+version: 1.0.0
+---
+
+# codegen-database
+
+Generate production-ready database code for schemas, DDL, OLTP queries, analytics SQL, migrations, indexes, stored procedures, and dialect-specific scripts. Use this as a router: classify the database artifact first, detect the dialect from context or repository evidence, then read only the relevant references.
+
+## Variant Detection
+
+**Route from artifact type before choosing syntax details.**
+
+- **Schema design:** Requests for entities, tables, relationships, normalization, constraints, ERD-to-DDL, or "what tables do I need" route to `references/schema-design.md`.
+- **Migrations:** Requests for up/down migrations, Flyway, Liquibase, Rails/ActiveRecord migrations, Alembic, Prisma migrations, rollback, data backfills, or deployment-safe DDL route to `references/migration.md`.
+- **OLTP SQL:** Requests for queries, DML, views, indexes, transactions, upserts, stored procedures, or query optimization route to `references/common.md`, then the dialect reference.
+- **Analytics SQL:** Requests for warehouses, metrics, cohorts, funnels, retention, partitioned fact tables, dbt-like transformations, BigQuery, Snowflake, or ClickHouse route to `references/analytics.md`, then any matching warehouse reference.
+- **Adjacent skills:** Use `report-db-health` for database health findings from existing telemetry. Use `strategy-backup` for backup and recovery policy. Use `writer-spec` for data contracts when the output is prose rather than executable database code.
+- **Ambiguity:** If the artifact type or database remains genuinely ambiguous after inspecting context, ask one short question naming the likely choices.
+
+## Dialect Routing
+
+**Choose the database from explicit signals, then repository evidence, then ask only if the choice changes the code.**
+
+| Signal | Reference |
+| --- | --- |
+| PostgreSQL, Postgres, `JSONB`, `TIMESTAMPTZ`, `ON CONFLICT`, `pg`, `psql` | `references/postgres.md` |
+| MySQL, MariaDB, `AUTO_INCREMENT`, `ON DUPLICATE KEY`, `mysql2` | `references/mysql.md` |
+| SQL Server, MSSQL, T-SQL, `pyodbc`, `OFFSET-FETCH`, `MERGE` | `references/mssql.md` |
+| SQLite, embedded/mobile/local database, FTS5, WAL mode | `references/sqlite.md` |
+| Oracle, PL/SQL, sequences, `DUAL`, Oracle hints | `references/oracle.md` |
+| BigQuery, GoogleSQL, ARRAY/STRUCT, partitioned warehouse tables | `references/bigquery.md` |
+| Snowflake, VARIANT, streams/tasks, clustering, time travel | `references/snowflake.md` |
+| ClickHouse, MergeTree, materialized views, sparse indexes | `references/clickhouse.md` |
+| CockroachDB, distributed SQL, follower reads, regional tables | `references/cockroachdb.md` |
+
+If the user asks for portable SQL, use `references/common.md` and avoid dialect-specific syntax unless you clearly mark alternatives.
+
+## Working Rules
+
+**Database code must be executable, reversible when applicable, and honest about assumptions.**
+
+- **Inspect first:** Read existing migrations, schema files, ORM models, query builders, naming conventions, fixtures, and migration tooling before editing repository files.
+- **Prefer structural guarantees:** Encode business rules with constraints, foreign keys, uniqueness, checks, generated columns, and transaction boundaries before relying on application-only enforcement.
+- **Keep migrations safe:** Make destructive DDL explicit, separate schema changes from risky data rewrites when needed, and include rollback or forward-fix guidance when true rollback is unsafe.
+- **Use parameters:** Never generate SQL that interpolates user input into executable statements. Use the placeholder style for the target dialect or framework.
+- **Index deliberately:** Tie each recommended index to a query, constraint, or access pattern. Avoid adding write-costly indexes without a reason.
+- **Handle concurrency:** Use transactions, locks, isolation levels, uniqueness, idempotency keys, or retry notes when the database operation can race.
+- **Respect dialect limits:** Do not mix syntax across engines. If the dialect is unknown and syntax materially differs, ask once instead of producing decorative nonsense in SQL clothing.
+- **Verify locally:** Run the narrowest relevant migration check, SQL parser, formatter, test, or application test available. If no database is available, state what was reviewed statically.
+
+## Implementation Flow
+
+**Move from existing schema to minimal database change to verification.**
+
+1. Identify artifact type and dialect, then read the selected artifact reference and dialect reference.
+2. Inspect the closest existing schema, migration, query, model, and tests.
+3. Plan the minimal database surface: tables, columns, constraints, indexes, migrations, queries, and tests or fixtures.
+4. Edit using project naming and migration conventions.
+5. Add or update focused tests, fixtures, or migration assertions when the repository supports them.
+6. Run focused verification and fix regressions within scope.
+
+## Output Format
+
+**Return runnable database code plus the operational context needed to use it.**
+
+When editing a repository, finish with changed files, commands run, and verification status.
+
+When drafting code only, include:
+
+```text
+Assumptions:
+- ...
+
+SQL / Migration:
+- ...
+
+Performance:
+- ...
+
+Rollback / Safety:
+- ...
+```