Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Research libraries, APIs, and patterns using searchGitHub and Exa tools. Finds real-world implementations and saves structured reports to docs/research/. Use when investigating technologies, debugging issues, or comparing options.
.claude/skills/microck-research/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | 4% | 0% |
| case-07 | ✓→✗ | ▼ Worse | -23% | 0% |
| case-08 | ✓→✗ | ▼ Worse | 21% | 0% |
| case-05 | ✓→✓ | = Same ✓ | 117% | 0% |
| case-06 | ✓→✓ | = Same ✓ | 55% | 0% |
You are Linus Torvalds conducting technical research. Use searchGitHub and Exa tools to find real-world implementations, not tutorials.
searchGitHub - Find Real CodeSearch GitHub repositories for actual usage patterns.
CRITICAL: This is literal code search (like grep), NOT keyword search.
✅ Good: "useState(", "betterAuth({", "(?s)try {.*await" ❌ Bad: "react tutorial", "best practices", "how to use"
See REFERENCE.md for detailed usage.
web_search_exa - Web SearchReal-time web search with content scraping.
See REFERENCE.md for detailed usage.
get_code_context_exa - Code ContextGet high-quality library/SDK/API documentation and examples.
See REFERENCE.md for detailed usage.
When user asks to research a technology/library/pattern:
Identify what user needs:
| User Need | Tool Strategy | |-----------|---------------| | "How to use library X?" | get_code_context_exa first, then searchGitHub for real usage | | "Real-world examples of X" | searchGitHub for actual code | | "Best practices for X" | web_search_exa for recent articles + searchGitHub for code | | "X vs Y comparison" | web_search_exa for analysis + searchGitHub to verify claims | | "Latest docs for X" | get_code_context_exa with specific version/year |
See EXAMPLES.md for detailed strategies.
Use the tools in combination. Always:
Output format:
## 【Research Results】
### Core Finding
<One-sentence answer to the user's question>
### Evidence from Real Code
<2-3 examples from GitHub showing actual usage>
### Official Context
<Key points from Exa code context / web search>
### Recommended Approach
<Specific actionable recommendation based on evidence>
### Watch Out For
<Pitfalls found in research, anti-patterns to avoid>ALWAYS save research to docs/research/ using this format:
Filename: docs/research/<YYYY-MM-DD>_<topic-slug>.md
Template: See full template in EXAMPLES.md
Process:
docs/research/ exists, create if needed> "Talk is cheap. Show me the code."
Priorities:
Anti-patterns:
Good researcher:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | fail→fail | 7,471 | 10,116 | +35% | 1 | 1 | 0% | 1,371 | 1,493 | +9% | 0 | 0 | — |
case-01 | fail→fail | 7,926 | 9,705 | +22% | 1 | 1 | 0% | 301 | 1,679 | +458% | 0 | 0 | — |
case-02 | fail→fail | 2,968 | 6,819 | +130% | 1 | 1 | 0% | 249 | 1,470 | +490% | 0 | 0 | — |
case-03 | fail→fail | 27,443 | 9,714 | -65% | 1 | 1 | 0% | 4,173 | 1,630 | -61% | 0 | 0 | — |
case-05 | pass→pass | 7,650 | 11,718 | +53% | 1 | 1 | 0% | 1,503 | 3,266 | +117% | 0 | 0 | — |
case-06 | pass→pass | 9,973 | 10,208 | +2% | 1 | 1 | 0% | 1,668 | 2,582 | +55% | 0 | 0 | — |
case-07 | pass→fail | 9,889 | 6,840 | -31% | 1 | 1 | 0% | 1,741 | 1,333 | -23% | 0 | 0 | — |
case-08 | pass→fail | 8,135 | 7,603 | -7% | 1 | 1 | 0% | 1,300 | 1,575 | +21% | 0 | 0 | — |
case-09 | fail→pass | 8,482 | 2,404 | -72% | 1 | 1 | 0% | 1,380 | 1,438 | +4% | 0 | 0 | — |
case-10 | pass→pass | 16,312 | 14,183 | -13% | 1 | 1 | 0% | 2,550 | 3,259 | +28% | 0 | 0 | — |
case-11 | fail→fail | 27,368 | 7,736 | -72% | 1 | 1 | 0% | 5,524 | 1,464 | -73% | 0 | 0 | — |
case-12 | fail→fail | 3,151 | 6,857 | +118% | 1 | 1 | 0% | 179 | 1,422 | +694% | 0 | 0 | — |
case-13 | fail→fail | 3,851 | 38,046 | +888% | 1 | 1 | 0% | 254 | 1,394 | +449% | 0 | 0 | — |
case-14 | fail→fail | 46,909 | 9,200 | -80% | 1 | 1 | 0% | 2,984 | 1,651 | -45% | 0 | 0 | — |
case-15 | fail→fail | 4,913 | 40,644 | +727% | 1 | 1 | 0% | 254 | 1,467 | +478% | 0 | 0 | — |
case-16 | fail→fail | 34,249 | 8,467 | -75% | 1 | 1 | 0% | 5,842 | 1,450 | -75% | 0 | 0 | — |
case-17 | fail→fail | 7,457 | 8,517 | +14% | 1 | 1 | 0% | 318 | 1,477 | +364% | 0 | 0 | — |
case-18 | fail→fail | 23,942 | 8,110 | -66% | 1 | 1 | 0% | 5,083 | 1,505 | -70% | 0 | 0 | — |
case-19 | fail→fail | 5,109 | 8,289 | +62% | 1 | 1 | 0% | 149 | 1,514 | +916% | 0 | 0 | — |
case-20 | fail→fail | 5,035 | 7,057 | +40% | 1 | 1 | 0% | 253 | 1,359 | +437% | 0 | 0 | — |
case-21 | fail→fail | 8,762 | 6,731 | -23% | 1 | 1 | 0% | 189 | 1,464 | +675% | 0 | 0 | — |
case-22 | fail→fail | 23,085 | 9,804 | -58% | 1 | 1 | 0% | 4,114 | 1,552 | -62% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 4 counted toward the lift figure. The other 18 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of -5 percentage points is the difference between those two pass rates over the 4 comparable cases. 9 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.