Search First

1---2name: search-first3description: Research-before-coding workflow. Search for existing tools, libraries, and patterns before writing custom code. Invokes the researcher agent.4origin: ECC5---6 7# /search-first — Research Before You Code8 9Systematizes the "search for existing solutions before implementing" workflow.10 11## Trigger12 13Use this skill when:14- Starting a new feature that likely has existing solutions15- Adding a dependency or integration16- The user asks "add X functionality" and you're about to write code17- Before creating a new utility, helper, or abstraction18 19## Workflow20 21```22┌─────────────────────────────────────────────┐23│  1. NEED ANALYSIS                           │24│     Define what functionality is needed      │25│     Identify language/framework constraints  │26├─────────────────────────────────────────────┤27│  2. PARALLEL SEARCH (researcher agent)      │28│     ┌──────────┐ ┌──────────┐ ┌──────────┐  │29│     │  npm /   │ │  MCP /   │ │  GitHub / │  │30│     │  PyPI    │ │  Skills  │ │  Web      │  │31│     └──────────┘ └──────────┘ └──────────┘  │32├─────────────────────────────────────────────┤33│  3. EVALUATE                                │34│     Score candidates (functionality, maint, │35│     community, docs, license, deps)         │36├─────────────────────────────────────────────┤37│  4. DECIDE                                  │38│     ┌─────────┐  ┌──────────┐  ┌─────────┐  │39│     │  Adopt  │  │  Extend  │  │  Build   │  │40│     │ as-is   │  │  /Wrap   │  │  Custom  │  │41│     └─────────┘  └──────────┘  └─────────┘  │42├─────────────────────────────────────────────┤43│  5. IMPLEMENT                               │44│     Install package / Configure MCP /       │45│     Write minimal custom code               │46└─────────────────────────────────────────────┘47```48 49## Decision Matrix50 51| Signal | Action |52|--------|--------|53| Exact match, well-maintained, MIT/Apache | **Adopt** — install and use directly |54| Partial match, good foundation | **Extend** — install + write thin wrapper |55| Multiple weak matches | **Compose** — combine 2-3 small packages |56| Nothing suitable found | **Build** — write custom, but informed by research |57 58## How to Use59 60### Quick Mode (inline)61 62Before writing a utility or adding functionality, mentally run through:63 640. Does this already exist in the repo? → `rg` through relevant modules/tests first651. Is this a common problem? → Search npm/PyPI662. Is there an MCP for this? → Check `~/.claude/settings.json` and search673. Is there a skill for this? → Check `~/.claude/skills/`684. Is there a GitHub implementation/template? → Run GitHub code search for maintained OSS before writing net-new code69 70### Full Mode (agent)71 72For non-trivial functionality, launch the researcher agent:73 74```75Task(subagent_type="general-purpose", prompt="76  Research existing tools for: [DESCRIPTION]77  Language/framework: [LANG]78  Constraints: [ANY]79 80  Search: npm/PyPI, MCP servers, Claude Code skills, GitHub81  Return: Structured comparison with recommendation82")83```84 85## Search Shortcuts by Category86 87### Development Tooling88- Linting → `eslint`, `ruff`, `textlint`, `markdownlint`89- Formatting → `prettier`, `black`, `gofmt`90- Testing → `jest`, `pytest`, `go test`91- Pre-commit → `husky`, `lint-staged`, `pre-commit`92 93### AI/LLM Integration94- Claude SDK → Context7 for latest docs95- Prompt management → Check MCP servers96- Document processing → `unstructured`, `pdfplumber`, `mammoth`97 98### Data & APIs99- HTTP clients → `httpx` (Python), `ky`/`got` (Node)100- Validation → `zod` (TS), `pydantic` (Python)101- Database → Check for MCP servers first102 103### Content & Publishing104- Markdown processing → `remark`, `unified`, `markdown-it`105- Image optimization → `sharp`, `imagemin`106 107## Integration Points108 109### With planner agent110The planner should invoke researcher before Phase 1 (Architecture Review):111- Researcher identifies available tools112- Planner incorporates them into the implementation plan113- Avoids "reinventing the wheel" in the plan114 115### With architect agent116The architect should consult researcher for:117- Technology stack decisions118- Integration pattern discovery119- Existing reference architectures120 121### With iterative-retrieval skill122Combine for progressive discovery:123- Cycle 1: Broad search (npm, PyPI, MCP)124- Cycle 2: Evaluate top candidates in detail125- Cycle 3: Test compatibility with project constraints126 127## Examples128 129### Example 1: "Add dead link checking"130```131Need: Check markdown files for broken links132Search: npm "markdown dead link checker"133Found: textlint-rule-no-dead-link (score: 9/10)134Action: ADOPT — npm install textlint-rule-no-dead-link135Result: Zero custom code, battle-tested solution136```137 138### Example 2: "Add HTTP client wrapper"139```140Need: Resilient HTTP client with retries and timeout handling141Search: npm "http client retry", PyPI "httpx retry"142Found: got (Node) with retry plugin, httpx (Python) with built-in retry143Action: ADOPT — use got/httpx directly with retry config144Result: Zero custom code, production-proven libraries145```146 147### Example 3: "Add config file linter"148```149Need: Validate project config files against a schema150Search: npm "config linter schema", "json schema validator cli"151Found: ajv-cli (score: 8/10)152Action: ADOPT + EXTEND — install ajv-cli, write project-specific schema153Result: 1 package + 1 schema file, no custom validation logic154```155 156## Anti-Patterns157 158- **Jumping to code**: Writing a utility without checking if one exists159- **Ignoring MCP**: Not checking if an MCP server already provides the capability160- **Over-customizing**: Wrapping a library so heavily it loses its benefits161- **Dependency bloat**: Installing a massive package for one small feature