Deep Research Skill: Evidence-Backed Reports for Your Agent

Where this fits

You're evaluating two vector databases for a new agent feature. You need current benchmarks, community sentiment, and API ergonomics — not a generic search summary. The deep-research skill turns that investigation into a structured findings.md report, so your agent can produce evidence-backed comparisons without you hand-curating sources.

Why agents benefit

• Produces a reproducible findings.md artifact that you can commit to your repo or feed into downstream decision-making.
• Uses Exa as the research backend, which returns high-quality, semantically relevant web results — better than a raw search for technical topics.
• Respects workflow conventions: it accepts an outputPath argument, so reports land where your project expects them.
• Automates the "research then summarize" loop, freeing your agent to focus on analysis and recommendations.
• Includes a fallback to general web search when Exa is unavailable, with explicit user approval — no silent degradation.

Practical scenarios

Architectural investigation

Before committing to a microservices vs. modular monolith approach, your agent runs deep-research on trade-offs, real-world case studies, and migration pain points. The resulting findings.md becomes the starting point for your design doc.

Vendor and library comparison

When choosing between React Query and SWR, the skill pulls recent blog posts, GitHub discussions, and release notes. It structures the evidence by criteria like bundle size, caching strategy, and maintainer activity — saving you hours of tab-hopping.

Technology choice validation

Your agent needs to confirm that a new database driver is stable enough for production. Deep-research gathers GitHub issues, Stack Overflow mentions, and version history, then flags red flags in the report.

Add it to your agent workflow

Install with:

npx skills add vanillagreencom/kendex --skill deep-research

Then invoke it in your agent harness (e.g., Claude Code, Codex, or Cursor) with:

scripts/deep-research "compare PostgreSQL vs. SQLite for a local-first app" --output findings.md

Set EXA_API_KEY in your environment. The skill will write a structured report to findings.md, ready for your next step.

Compare before adopting

Before you rely on this skill, check three things. First, quality signals: does the report cite primary sources and dates? Second, maintenance freshness: the repo has 63 stars and an active issue tracker — but verify the last commit and Exa API changes. Third, alternatives: you might already have a research tool like web_research in your harness; compare output structure and reliability. Also assess workflow fit: does findings.md align with your existing documentation patterns? For a deeper look, see the skill page on OpenAgentSkill.

Why it is worth tracking

The skill is young but promising: 63 stars, an MIT license, and a clear dependency on Exa — which is a strong signal for research-focused agents. The author actively maintains it, and the kendex repo bundles related skills. Track it if you frequently need evidence-backed research artifacts; evaluate it when your agent's research output needs to be more than a chat reply. As the skill matures, expect better error handling and more configurable report formats.