external-context
The external-context skill decomposes complex queries into two to five independent search facets, then spawns parallel document-specialist Claude agents to fetch relevant external documentation, web references, and technical context. Use this skill when you need comprehensive, sourced information about technical topics like best practices, library comparisons, or current framework patterns that benefit from multiple specialized research angles conducted simultaneously.
git clone --depth 1 https://github.com/Yeachan-Heo/oh-my-claudecode /tmp/external-context && cp -r /tmp/external-context/skills/external-context ~/.claude/skills/external-contextSKILL.md
# External Context Skill Fetch external documentation, references, and context for a query. Decomposes into 2-5 facets and spawns parallel document-specialist Claude agents. ## Usage ``` /oh-my-claudecode:external-context <topic or question> ``` ### Examples ``` /oh-my-claudecode:external-context What are the best practices for JWT token rotation in Node.js? /oh-my-claudecode:external-context Compare Prisma vs Drizzle ORM for PostgreSQL /oh-my-claudecode:external-context Latest React Server Components patterns and conventions ``` ## Protocol ### Step 1: Facet Decomposition Given a query, decompose into 2-5 independent search facets: ```markdown ## Search Decomposition **Query:** <original query> ### Facet 1: <facet-name> - **Search focus:** What to search for - **Sources:** Official docs, GitHub, blogs, etc. ### Facet 2: <facet-name> ... ``` ### Step 2: Parallel Agent Invocation Fire independent facets in parallel via Task tool: ``` Task(subagent_type="oh-my-claudecode:document-specialist", model="sonnet", prompt="Search for: <facet 1 description>. Use WebSearch and WebFetch to find official documentation and examples. Cite all sources with URLs.") Task(subagent_type="oh-my-claudecode:document-specialist", model="sonnet", prompt="Search for: <facet 2 description>. Use WebSearch and WebFetch to find official documentation and examples. Cite all sources with URLs.") ``` Maximum 5 parallel document-specialist agents. ### Step 3: Synthesis Output Format Present synthesized results in this format: ```markdown ## External Context: <query> ### Key Findings 1. **<finding>** - Source: [title](url) 2. **<finding>** - Source: [title](url) ### Detailed Results #### Facet 1: <name> <aggregated findings with citations> #### Facet 2: <name> <aggregated findings with citations> ### Sources - [Source 1](url) - [Source 2](url) ``` ## Configuration - Maximum 5 parallel document-specialist agents - No magic keyword trigger - explicit invocation only
Clean AI-generated code slop with a regression-safe, deletion-first workflow and optional reviewer-only mode
Pre-planning consultant for requirements analysis (Opus)
Strategic Architecture & Debugging Advisor (Opus, READ-ONLY)
Process-first advisor routing for Claude, Codex, Gemini, Antigravity, Grok, or Cursor via `omc ask`, with artifact capture and no raw CLI assembly
Full autonomous execution from idea to working code
Stateful single-mission improvement loop with strict evaluator contract, markdown decision logs, and max-runtime stop behavior
Cancel any active OMC mode (autopilot, ralph, ultragoal, swarm, ultrapilot, pipeline, team) and clean up retired legacy state
Claude-Codex-Gemini tri-model orchestration via /ask codex + /ask antigravity (or gemini), then Claude synthesizes results