From yellow-research
Reference conventions for yellow-research plugin — slug naming, output format, source selection, API key setup, graceful degradation, and when to compound findings.
npx claudepluginhub kinginyellows/yellow-plugins --plugin yellow-researchThis skill uses the workspace's default tool permissions.
Convert the research topic to a slug for the output filename:
Researches topics comprehensively with web search, scraping, and AI agents, producing timestamped Markdown documentation in .claude/research/.
Conducts multi-source research on topics via web search, documentation queries, matrix memory, and GitHub using a 5-phase pipeline. Outputs markdown reports at quick/standard/exhaustive depths.
Conducts systematic internet research via strategic questioning, multi-source analysis, credibility evaluation with tier system, and structured .research/ reports. Use for best practices, comparisons, technology evaluations, trends.
Share bugs, ideas, or general feedback.
Convert the research topic to a slug for the output filename:
[a-z0-9-] only — lowercase, hyphens, no spaces or special charsreact-server-components-2026exa-neural-searchnotion-vs-linear-competitordocs/research/<slug>.md exists, append -2, -3All deep research saves to: docs/research/<slug>.md
Create the directory if it doesn't exist:
mkdir -p docs/research
# <Topic Title>
**Date:** YYYY-MM-DD
**Sources:** Perplexity, EXA, Tavily (list which were used)
## Summary
[2-3 sentence executive summary]
## Key Findings
### <Subtopic 1>
[Findings with source citations]
### <Subtopic 2>
[Findings with source citations]
## Sources
- [Source Title](URL) — brief note on what was found here
| Query Type | Primary Source | Secondary |
|---|---|---|
| Library / framework docs | Context7 | EXA get_code_context_exa |
| Code examples, GitHub patterns | EXA get_code_context_exa | GitHub grep |
| Recent news, current events | Perplexity perplexity_search | Tavily tavily_search |
| Competitive / company research | EXA company_research_exa | Perplexity |
| Deep technical report | Perplexity perplexity_research | Tavily tavily_research |
| AST / structural code patterns | ast-grep find_code | ast-grep find_code_by_rule |
| Long-horizon async report | Parallel createDeepResearch | EXA deep_researcher_start |
| Specific URL content | EXA crawling_exa | Tavily tavily_extract |
After /research:deep, run /compound if the findings include:
Routine informational research does not need to be compounded.
If an MCP server is unavailable (key not set, connection error, rate limit):
Add to ~/.zshrc (or equivalent shell config):
export EXA_API_KEY="your-key-here"
export TAVILY_API_KEY="your-key-here"
export PERPLEXITY_API_KEY="your-key-here"
Get keys from:
The Parallel Task server uses OAuth (no API key needed). Claude Code handles authentication automatically — you'll be prompted to authorize on first use.
After adding keys: source ~/.zshrc and restart Claude Code.
After installing yellow-research, always verify actual tool names before writing agent code or troubleshooting:
ToolSearch keyword: "exa"
Tool name: mcp__plugin_yellow-research_exa__...
ToolSearch keyword: "tavily"
Tool name: mcp__plugin_yellow-research_tavily__...
ToolSearch keyword: "perplexity"
Tool name: mcp__plugin_yellow-research_perplexity__...
ToolSearch keyword: "parallel"
Tool name: mcp__plugin_yellow-research_parallel__...
ToolSearch keyword: "ast-grep__find_code"
Tool name: mcp__plugin_yellow-research_ast-grep__find_code
ToolSearch keyword: "ast-grep__find_code_by_rule"
Tool name: mcp__plugin_yellow-research_ast-grep__find_code_by_rule
ToolSearch keyword: "ast-grep__dump_syntax_tree"
Tool name: mcp__plugin_yellow-research_ast-grep__dump_syntax_tree
ToolSearch keyword: "ast-grep__test_match_code_rule"
Tool name: mcp__plugin_yellow-research_ast-grep__test_match_code_rule
Naming convention: mcp__plugin_yellow-research_<server>__<tool>.
Or run /mcp in Claude Code to see all registered tools.
Never trust LLM-generated tool names — they may be fictitious. Empirical
verification is required. If names differ from what's in agents/commands, update
the allowed-tools lists and any MCP tool references in agent bodies.