From r5
This skill should be used when the user asks to "research a GitHub repository", "investigate org/repo", "deepwiki research", "analyze repository implementation", "understand how X repo works", or provides an "owner/repo" format GitHub repository name for investigation. Use for researching any GitHub repository's implementation, codebase structure, or internal workings using Deepwiki MCP.
How this skill is triggered — by the user, by Claude, or both
Slash command
/r5:deepwiki-researchThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Investigate any GitHub repository's implementation details using Deepwiki MCP. This skill enables research on undocumented internal workflows, execution patterns, and core functionalities across any public GitHub repository.
Investigate any GitHub repository's implementation details using Deepwiki MCP. This skill enables research on undocumented internal workflows, execution patterns, and core functionalities across any public GitHub repository.
This plugin automatically configures Deepwiki MCP for public repository research. No manual setup required.
For private repositories:
If you need to research private repositories, update plugins/r5/.mcp.json to include your API key:
{
"deepwiki": {
"url": "https://mcp.devin.ai/mcp",
"headers": {
"Authorization": "Bearer ${DEEPWIKI_API_KEY}"
}
}
}
Then set the environment variable:
export DEEPWIKI_API_KEY="your-api-key-here"
See Deepwiki MCP documentation for API key setup.
Use this skill when user mentions any of the following:
Extract the repository name from user input. The format should be owner/repo (e.g., facebook/react, python/cpython, Comfy-Org/ComfyUI).
If user doesn't specify a repository, ask: "Which GitHub repository would you like to investigate? Please provide it in owner/repo format."
First, review the existing conversation history. The user may have already explained their purpose in prior messages — extract both the topic and the goal from that context without asking again.
If the goal is still unclear after reviewing the conversation, ask one focused question before proceeding:
"What are you ultimately trying to do with this? (e.g., understand how it works, find where it's implemented, debug an issue, implement something similar)"
Only proceed to Step 3 once you understand both:
3 tools, use in order — this sequence dramatically improves precision:
Step 3a: Get topic list (always start here for unfamiliar repos)
read_wiki_structure(repoName="owner/repo")
# → Returns list of wiki topics (e.g., "Architecture", "Authentication", "Data Flow")
Step 3b: Read the most relevant topic in full (this is the key step)
read_wiki_contents(repoName="owner/repo", topic="Architecture")
# → Returns comprehensive wiki page: accurate file paths, class names, data flow diagrams
# → Far more reliable than ask_question alone — use this before asking questions
Step 3c: Ask follow-up questions (for specifics not covered in the wiki page)
ask_question(
repoName="owner/repo",
question="Your specific question about the implementation"
)
Why this order matters:
read_wiki_contentsreturns the full curated wiki page for a topic, giving far more accurate file paths and implementation details than RAG-based Q&A alone. Use it beforeask_questionto ground your follow-up questions in real context.
See references/deepwiki-tools.md for detailed API reference.
Extract specific file paths mentioned in the response (e.g., src/index.js, lib/core.py).
Always verify Deepwiki information by viewing the actual source code. Deepwiki may be outdated or incorrect.
# Step 5a: View the file identified in Step 4
gh api repos/owner/repo/contents/path/to/file.py | jq -r '.content' | base64 -d
# Step 5b: If you need to search for specific functions
gh api search/code -X GET -f q="repo:owner/repo function_name" | jq '.items[].path'
# Step 5c: View raw file directly (alternative)
curl -s https://raw.githubusercontent.com/owner/repo/main/path/to/file.py
Compare what Deepwiki provided with the actual source code:
If the source code reveals related functions or files, repeat from Step 3.
gh api repos/owner/repo/commits -f path=path/to/file.py -f per_page=3
npx claudepluginhub radius5-study/r5-plugin --plugin r5Conducts multi-turn iterative deep research on codebase topics by tracing code paths across files, mapping architecture, and analyzing data flows with strict evidence standards.
Retrieves code context summaries, API docs, and examples from GitHub repos, npm/pip libraries, code searches, and clones using isolated agents.
Clones a GitHub repo and sets up an interactive study session for reading and analyzing the codebase. Answers questions with source-of-truth evidence from the actual code.