# Documentation Lookup > Resolve library/API behavior via Context7 MCP and primary vendor docs before writing code Use when this capability is needed. - Skill: `tomevault-io/documentation-lookup-15` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/documentation-lookup-15` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/documentation-lookup-15/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/documentation-lookup-15 --- # documentation-lookup Before writing code that touches an external library, framework, or API — look it up. Do not guess API shapes from training data; they drift. ## Lookup priority 1. **GitHub code search** (`gh search code`, `gh search repos`) — find real usage in the wild 2. **Context7 MCP** (`mcp__context7__resolve-library-id`, then `mcp__context7__query-docs`) — curated, versioned docs 3. **Primary vendor docs** (direct URL from the package's `repository` / `homepage` field in its registry listing) 4. **Exa** — broader web search, only when 1–3 don't answer the question ## Packages this repo will pull in (PRs 2–6) | Package | Why | Docs entry point | | ------------------------------------ | --------------------- | -------------------------------------------------------------------------- | | prettier | Formatter | https://prettier.io/docs/en/ | | eslint, @eslint/js | JS lint (flat config) | https://eslint.org/docs/latest/ | | stylelint, stylelint-config-standard | CSS lint | https://stylelint.io/ | | html-validate | HTML lint | https://html-validate.org/ | | @playwright/test | E2E | https://playwright.dev/docs/intro | | @axe-core/playwright | A11y scan | https://github.com/dequelabs/axe-core-npm/tree/develop/packages/playwright | | sharp-cli (or Squoosh CLI) | Image pipeline | https://github.com/vseventer/sharp-cli | ## Prompts that work - `mcp__context7__resolve-library-id "@playwright/test"` → get stable ID - `mcp__context7__query-docs` with a specific question: _"How do I configure projects for desktop and mobile viewport in playwright.config.ts?"_ Ask narrow questions. Broad ones ("how does Playwright work") waste tokens and return boilerplate. ## What to record when you look something up If the finding is non-obvious and likely to recur, capture it: - **Behavior-of-the-tool fact** → inline code comment (one short line, the WHY) - **Project decision based on the finding** → `tasks/plan.md` or `tasks/lessons.md` - **Version-specific quirk** → note in the relevant skill's SKILL.md ## What NOT to do - Don't paste full docs into the conversation — link or cite the section - Don't trust unversioned blog posts over the vendor's release notes - Don't refactor code based on a Stack Overflow answer without checking the vendor doc - Don't invent option names, flag names, or API shapes --- > Source: [danielpesa7/landing-page-2-dato](https://github.com/danielpesa7/landing-page-2-dato) — distributed by [TomeVault](https://tomevault.io).