team-attention/hoyeon/reference-seek
Claude Code and agent skill by team-attention/hoyeon
- community
Checking that the SKILL.md file is still in the repository.
This skill should be used when the user asks to "find references", "참고할 만한 거", "similar implementation", "오픈소스 있나", "how others do this", "레퍼런스 찾아줘", or needs to find existing patterns (internal) and open-source examples (external) for implementing a feature.
Install
npx vskill@latest install team-attention/hoyeon/reference-seekSKILL.md
Reference Seek - Find Implementation References (v2.0.0)
Find internal patterns, quality-filtered GitHub repos with code-level insights, official docs, and external examples.
Purpose
When building a feature, find what you can reuse or learn from:
- Internal: Existing patterns in the codebase
- GitHub: Quality-filtered repos with actual code excerpts
- Official Docs: context7-powered library documentation
- External: Blog posts, tutorials, best practices
Runtime Surface
Claude Code
- Use
Task(Explore),gh, context7 MCP when available, andWebSearchas
described below.
Codex
- Map internal codebase search to
hoyeon-code-explorerwhen installed. - Use Bash-first
gh apiandcurlfor GitHub discovery and code excerpts. - Do not require context7 MCP for v1; use it only if already available in the
current runtime. Otherwise, use official docs via the available Codex web search/fetch surface and label that fallback.
- If no native subagent adapter is loaded, perform the internal search directly
with rg and targeted file reads.
Use Cases
- "Looking to implement OAuth login — any good references?"
- "Find references for implementing a rate limiter"
- "How do I implement pagination? Find me some references"
- "Any open source projects similar to WebSocket connection management?"
Execution
Step 0: Dependency Check + Topic Classification (1 message)
Run in parallel:
0-A. gh auth check:
| Result | Variable | |
|---|---|---|
| Authenticated | GH_AVAILABLE=true | |
| Not authenticated / gh not found | GH_AVAILABLE=false |
0-B. Topic extraction + classification (reasoning, same message):
Extract from user request:
TOPIC: Normalized English topic (e.g., "rate limiting", "OAuth authentication")TOPIC_CLASS:library|pattern|conceptGH_QUERY: GitHub search query (e.g.,rate+limiter,oauth+middleware)GH_LANG_FILTER: Language filter inferred from codebase (e.g.,+language:typescript)CONTEXT7_LIB: Library name for context7 resolve (iflibraryclass, use directly; otherwise best guess or skip)
Examples:
- "any rate limiter references?" → TOPIC:
rate limiting, CLASS:pattern, GH_QUERY:rate+limiter+middleware, CONTEXT7_LIB:rate-limiter-flexible - "NextAuth implementation references" → TOPIC:
NextAuth authentication, CLASS:library, GH_QUERY:nextauth+authentication, CONTEXT7_LIB:next-auth - "pagination references" → TOPIC:
pagination implementation, CLASS:pattern, GH_QUERY:pagination+cursor, CONTEXT7_LIB: skip
Step 1: Parallel Search (1 message, up to 4 sources)
CRITICAL: Run all available tracks in one message in parallel.
| Track | Tool | Purpose | |
|---|---|---|---|
| 1-A | Task(Explore) | Internal codebase pattern search | |
| 1-B | Bash: gh api search/repositories | GitHub repo search (stars-sorted, top 10) | |
| 1-C | mcp__context7__resolve-library-id | Library docs ID lookup | |
| 1-D | WebSearch | Blog/tutorial search (1 query only) |
1-A. Internal Search (Explore agent):
1-B. GitHub Search (only if GH_AVAILABLE=true):
1-C. context7 resolve (if CONTEXT7_LIB is set):
1-D. WebSearch (1 query only):
When GH_AVAILABLE=false: Skip 1-B. Run WebSearch 3 times instead (v1.0 fallback):
Add warning at top of output: > ⚠️ GitHub API unavailable (gh not authenticated). Results from web search only.
Step 2: Quality Filter + context7 Query (sequential after Step 1)
2-A. Quality Filter (reasoning, applied to 1-B results):
| Criteria | Threshold | |
|---|---|---|
| Stars | >= 100 (relax to >= 50 if fewer than 2 pass) | |
| Last push | Within 24 months | |
| License | MIT, Apache-2.0, BSD-2-Clause, BSD-3-Clause, ISC, MPL-2.0 | |
| Archived | Excluded (already filtered by jq) |
→ Top 5 passing repos = QUALIFIED_REPOS → Top 1-2 repos = DEEP_DIVE_TARGETS
If all repos filtered out: relax stars to >= 50. If still none, take top 3 with warning badge ⚠️ Below quality threshold.
2-B. context7 docs query (only if 1-C resolve succeeded):
If resolve failed in 1-C, skip this step entirely.
Step 3: Repo Deep Dive (sequential after Step 2)
For each repo in DEEP_DIVE_TARGETS (1-2 repos):
3-1. File tree lookup:
3-2. File selection (reasoning):
- Match files by topic keywords
- Prioritize: entry points, types/interfaces, core logic
- Skip files > 20KB
- Select max 3 files per repo
3-3. File content fetch:
3-4. Code analysis (reasoning): Extract from each file:
- Key data structures
- Algorithm / approach
- Configuration options
- Notable patterns worth adopting
If file tree or fetch fails for a repo, skip its deep dive and move to the next target.
Step 4: Synthesize & Present
Combine all sources into the output format below.
Output Format
Error Handling
| Situation | Response | |
|---|---|---|
| gh not authenticated | Skip 1-B, use WebSearch 3x fallback, add warning at top | |
| gh rate limit (403) | Fallback: gh search repos "{GH_QUERY}" --sort stars --limit 10 --json ... (different rate limit bucket) | |
| GH results 0 | Remove language filter, retry. If still 0, show "No GitHub results found" | |
| All repos filtered out | Relax stars to >= 50. If still none, show top 3 with warning badge | |
| context7 resolve fails | Omit "Official Docs" section entirely | |
| File tree / fetch fails | Skip that repo's deep dive, proceed to next target | |
| Explore returns nothing | Show "No existing patterns found in codebase" | |
| Topic too vague | Ask user for clarification before proceeding |
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/reference-seek/SKILL.md
Popularity
- Repository stars
- 115
- Repository forks
- 13
- 7-day trend score
- 24
Stars and forks are for the whole GitHub repository, not this skill alone. Refreshed 20 Mar 2026, 7 months ago.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/team-attention/hoyeon/reference-seek)