Universal web search across 100+ search engines and result types.
Documentation-only skill package for AI coding agents. No executable code — the deliverable is Markdown files that agents consume. Powered by SerpApi REST API via serpapi_search MCP tool, the serpapi CLI (preferred), official SDKs, or curl.
./
├── AGENTS.md # This file — skill discovery + project knowledge
├── README.md # GitHub landing page, installation instructions
├── LICENSE # MIT
├── docs/
│ └── api-key-setup.md # SERPAPI_KEY config per agent + CI/CD
└── skills/serpapi-web-search/
├── SKILL.md # Core skill definition (frontmatter + usage)
├── serpapi.yaml # Network policy preset
└── rules/
├── ENGINES.md # Full catalog of 107 search engines
├── examples.md # CLI examples for common search types
├── parameters.md # All query parameters with examples
├── response.md # Response format and result key reference
└── sdks.md # SDK quickstart: Python, JS, Go, Ruby, PHP, Java, .NET
Hidden (not tracked in git):
.opencode/— OpenCode plugin packaging (node_modules, duplicate SKILL.md).sisyphus/— Build/QA evidence and plans
| Task | Location | Notes |
|---|---|---|
| Add SDK quickstart | skills/serpapi-web-search/rules/sdks.md |
Python, JS, Go, Ruby, PHP, Java, .NET |
| Edit skill behavior/examples | skills/serpapi-web-search/SKILL.md |
Canonical agent-facing artifact |
| Add/update search engines | skills/serpapi-web-search/rules/ENGINES.md |
107 engines in categorized tables |
| Change API key instructions | docs/api-key-setup.md |
Per-agent setup (Claude Code, Cursor, etc.) |
| Update install instructions | README.md |
7 agent platforms + universal curl |
| Skill discovery metadata | AGENTS.md (this file) |
<available-skills> XML block |
- Default engine:
google_light— always recommend Light endpoints first for speed/cost. - API key placeholder: Use
your_key_hereconsistently (never hardcode real keys). - Env var name:
SERPAPI_KEY— standardized across all docs and examples. - Invocation order: MCP tool (
serpapi_search) → serpapi-cli → SDK (if writing code) → curl (last resort). - serpapi-cli agent tips: Use
--fieldsfor server-side filtering (reduces API payload),--jqfor client-side filtering. Combine both for minimum token usage. - curl examples: All use
${SERPAPI_KEY}variable reference. - SKILL.md frontmatter: YAML block with
name,description,licensefields. - No executable code: This repo is pure Markdown. No scripts, no tests, no build step.
- Never commit real API keys or hex strings that look like keys.
- Never reference competitor products (Brave, etc.) in any file.
- Never add executable code — this is a docs-only skill package.
- Light vs Full: Always default to Light engines. Only suggest full engine when user needs knowledge graph, local pack, or featured snippets.
No build/test/lint commands. Pure documentation repo.
# Verify no keys leaked
grep -rE "[a-f0-9]{32,}" --include="*.md" .
# Validate internal links exist
grep -ohE '\(([^)]+\.md[^)]*)\)' *.md docs/*.md skills/**/*.md | tr -d '()' | sort -u- macOS gotcha:
grep -P(Perl regex) unavailable on Darwin — usegrep -Eorsedinstead. - Link paths in AGENTS.md: Must be relative to repo root (not to the file's parent). Previously had a broken
references/ENGINES.mdlink — fixed toskills/serpapi-web-search/rules/ENGINES.md. - OpenCode packaging:
.opencode/contains a bundled copy of the skill +@opencode-ai/plugindependency. This is separate from the canonicalskills/directory. - Git history: 5 atomic commits on
master. No branches, no CI pipeline. - Submissions: Skill is publishable to agentskills.guide, SkillMD.ai, skills.rest, and ClawHub. See
.sisyphus/evidence/task-11-submissions.mdfor submission details.