smith-research
Proactive research protocol for version queries, APIs, and best practices. Use when answering questions about library versions, API documentation, or technology assessments. Covers research triggers, source citation, and confidence indicators.
When & Why to Use This Skill
This Claude skill implements a proactive research protocol designed to provide accurate, up-to-date information on software versions, API documentation, and technical best practices. By prioritizing official sources and real-time data fetching, it eliminates hallucinations and ensures developers receive verified technical guidance with proper citations and confidence indicators.
Use Cases
- Verifying the latest stable release and breaking changes of libraries (e.g., React, Next.js) to ensure project compatibility and avoid deprecated features.
- Retrieving real-time API endpoint specifications and usage examples from official documentation for unfamiliar third-party services.
- Conducting technology assessments to compare frameworks based on current-year industry standards and community best practices.
- Bypassing paywalls on technical platforms like Medium to access deep-dive tutorials and troubleshooting guides using specialized mirrors.
- Validating technical implementation details when error messages suggest version mismatches or outdated syntax.
| name | smith-research |
|---|---|
| description | Proactive research protocol for version queries, APIs, and best practices. Use when answering questions about library versions, API documentation, or technology assessments. Covers research triggers, source citation, and confidence indicators. |
Proactive Research Protocol
- Load if: Queries about versions, APIs, libraries, best practices
- Prerequisites: @smith-guidance/SKILL.md
CRITICAL: Mandatory Research Triggers (Primacy Zone)
Temporal constraint: ALWAYS get today's real date first as critical context for research queries.
MUST research when:
- Version/release queries ("latest version of X")
- API/documentation queries ("how to use X API")
- Best practices queries ("recommended way in [current year]")
- Technology assessment ("should I use X or Y")
- Unfamiliar technology (not in training data)
Research Methods (Priority Order)
Official docs - Most authoritative
fetch https://docs.python.org/3/library/asyncio.html
Package registry - Version info
- npm:
https://registry.npmjs.org/[package]/latest - PyPI:
https://pypi.org/pypi/[package]/json - GitHub:
https://api.github.com/repos/[owner]/[repo]/releases/latest
- npm:
Web search - Broad queries
- Include current year: "Next.js 15 best practices [current year]"
GitHub repo - Source of truth
- README, CHANGELOG, release notes
Source Citation
Always include:
- Source URL
- Date of retrieval
- Version referenced
Format:
"React 19 introduced `feature_name` [1]"
[1] Source Name: URL (retrieved YYYY-MM-DD)
Confidence indicators:
- High: "Per official documentation..."
- Medium: "Based on recent community discussion..."
- Low: "My training data suggests X, but couldn't verify..."
Paywall Workarounds
Medium.com: Replace domain with freedium-mirror.cfd
- Pattern:
https://freedium-mirror.cfd/https://medium.com/@author/article - Use when WebFetch returns paywall/403 errors
When NOT to Research
- Fundamental concepts (won't change)
- Opinion/analysis requests
- User provides current info themselves
Proactive Behavior
Research when:
- Answering would require guessing
- User's code references unfamiliar library
- Error suggests version incompatibility
- User mentions "latest", "current", "2024", "2025"
Inform user:
- "I checked the current documentation..."
- "I couldn't fetch latest info, but based on training..."
- Claiming certainty about versions without verification
- Providing outdated API examples when docs accessible
- Skipping research when accuracy matters
- @smith-guidance/SKILL.md - Honest principle
@smith-tools/SKILL.md- MCP fetch configuration
ACTION (Recency Zone)
Before answering version/API questions:
- Check if research trigger applies
- Fetch official docs or registry
- Cite source with URL and date
- Flag if info conflicts with training data