feat: add new skills, agent personas, and LLM integration checklist

[rickrapozo]

Summary

Adds three new skills, two agent personas, and one reference checklist to expand the project's coverage of common engineering workflows.

New Skills

• llm-api-integration -- Structured process for integrating LLM APIs (prompt safety, JSON schema output, exponential backoff, mock-based testing)
• • accessibility-testing-and-remediation -- End-to-end a11y audit workflow (automated checks, keyboard audit, screen reader validation, regression testing)
• • database-migrations-and-schema-design -- Schema design with constraints, strategic indexing, zero-downtime migrations (Expand-Contract pattern), and rollback scripts

New Personas

• a11y-auditor -- Accessibility Engineer persona for WCAG 2.1 AA audits
• • db-architect -- Database Architect persona for schema review, migration safety, and query optimization

New Reference

• references/llm-integration-checklist.md -- Quick-reference checklist for safe, resilient LLM integrations

Validation

• node scripts/validate-skills.js -> 27 skills, 0 errors, 0 warnings - PASSED
• • bash hooks/session-start-test.sh -> session-start JSON payload OK
    All new content follows the skill-anatomy.md format and agents/README.md guidelines.

[nucliweb]

Thanks for this, there's genuinely useful content here. One piece of feedback on packaging that I think would help it land.

Right now the PR bundles a lot into one changeset: 3 skills (llm-api-integration, accessibility-testing-and-remediation, database-migrations-and-schema-design), 2 personas (a11y-auditor, db-architect), and 1 reference checklist, across 762 lines and 7 files. That breadth makes it hard to review as a unit, and it overlaps with other open PRs that target the same areas independently:

• accessibility-testing-and-remediation overlaps with feat: add accessibility-engineering skill #261 (accessibility-engineering)
• llm-api-integration sits alongside the LLM cluster (Add llm-cost-optimization skill #287 cost, Add evaluating-llm-output skill #286 evals, Add reliable-agent-loops skill #285 agent loops, Add prompt-injection-defense skill (OWASP LLM01) #129 prompt-injection)

Would you be open to splitting this into atomic PRs, one skill per PR? That lets each piece be reviewed and merged on its own merits, and lets the maintainer coordinate the overlapping ones without blocking the rest.

In particular, database-migrations-and-schema-design looks like the strongest candidate to land standalone: it fills a real gap (no current skill covers schema design or migrations), it doesn't overlap with any other open PR, and the content is solid as-is, Expand-Contract for zero-downtime, strategic indexing, rollback scripts, and EXPLAIN ANALYZE profiling. Pulling just that one into its own PR would likely move fast.

Happy to help review the split PRs if useful.