feat: enhance algolia-mcp skill with evals and improved docs 🔬

[leonardocuco]

Summary

• Added evaluation framework for the algolia-mcp skill with 5 test scenarios comparing with-skill vs without-skill baseline
• Surfaced critical reference content into main SKILL.md for more robust behavior (filter syntax, clickAnalytics, thresholds, interpretation benchmarks)
• Added common workflows (Search Quality Audit, Recommendation Setup Check)

Key improvements to SKILL.md

• Search Filter Syntax section — facetFilters OR/AND arrays, numericFilters strings, Unix timestamp date filtering, attribute selection
• Analytics Key Details section — clickAnalytics: true guidance with which tools support it, data delay warning
• Interpreting Results section — no-results rate benchmarks, click position patterns, low CTR diagnosis
• Recommendation Thresholds section — 50/60/75 guidance, model parameter requirements table
• Common Workflows — Search Quality Audit (6-step diagnosis), Recommendation Setup Check
• algolia-cli cross-reference for write operations

Eval results

Eval	Without Skill	With Skill
Search with filters	17%	100%
Analytics report	33%	100%
Recommendations	33%	100%
Multi-step investigation	17%	100%
Date filtering + pagination	33%	100%
Average	27%	100%

The skill's value is overwhelmingly about MCP tool interface knowledge — every baseline fabricated wrong tool names, skipped the discovery step, and missed clickAnalytics: true.

Test plan

• Ran 5 eval scenarios with-skill and without-skill
• Verified 100% pass rate across all with-skill runs
• Confirmed baselines fail on tool names, discovery workflow, and parameter conventions
• Manual review of SKILL.md readability and completeness

🤖 Generated with Claude Code