← ClaudeAtlas

aio-discoverlisted

Find code, locate implementations, and understand how features work via parallel Explore agents. Step 1 of the codeflow discover → map → plan trio used before implementation.
aiocean/claude-plugins · ★ 3 · AI & Automation · score 65
Install: claude install-skill aiocean/claude-plugins
# Discover — Find Relevant Code ## Environment - GitNexus: !`npx gitnexus status 2>/dev/null && echo "AVAILABLE" || echo "NOT INSTALLED"` Hybrid search across the entire codebase using GitNexus. Use BEFORE planning or coding to understand what exists. ## Prerequisites - GitNexus indexed in project — run `npx gitnexus analyze` if not yet indexed - Verify with `npx gitnexus status` ## Workflow ### Step 1: Formulate 3-5 search queries From the user's request, generate diverse queries covering different angles: - **Functional**: "how does [feature] work" - **Structural**: "[component type] for [domain]" - **Cross-cutting**: "[pattern] across frontend and backend" - **Vietnamese OK**: GitNexus hybrid search handles multilingual ### Step 2: Run searches in parallel Use the GitNexus MCP `query` tool for each search: ``` query("query here") ``` Run 3-5 searches as separate parallel MCP tool calls. Alternatively use the CLI: ```bash npx gitnexus analyze ``` ### Step 3: Score and filter | Similarity | Relevance | |---|---| | >0.65 | Highly relevant — must read | | 0.55–0.65 | Related — worth knowing | | <0.55 | Tangential — skip unless desperate | ### Step 4: Enrich with GitNexus context (parallel) For each highly relevant file found, get symbol overview: ``` context(file) ``` This adds function names and signatures without reading the full file. Run in parallel for all relevant files. Optionally, get full context for the most important symbol: ``` context(file,