SandBase Skills

Browse the flagship Skill: skills.sh · Agentic Awesome Skills · Agent Skill Exchange · askill · Awesome Skills · skills.re · SkillsCat · OpenAgentSkill · AgentSkill.sh

Have a repeatable workflow that is missing? Request a Skill with a public example and acceptance criteria.

English | 中文 | 日本語 | 한국어 | Español | Français | Deutsch | Português

88 installable Agent Skills for research, social intelligence, marketing, and business workflows. Install into any compatible agent (DeepSeek Harness, Claude Code, Codex, Cursor, Gemini CLI) and start working immediately. The flagship research Skill works with host-provided search tools; connect SandBase when you want broader provider coverage.

SandBase also exposes 2,000+ models and APIs, including a unified surface for LLM, image, and video generation. Skills define the workflow; the SandBase API or MCP bridge supplies optional model, media, search, social, and data capabilities when a workflow needs them.

Start with multi-source-search: it runs with your agent's existing search tools, ships a worked evidence-ledger example, and includes an offline validator. If it improves a real workflow, star the repository so other builders can discover it.

What are Skills?

A Skill is an instruction file that teaches an AI agent how to do one specific job. Each Skill defines a repeatable workflow, evidence rules, and output format. Portable Skills can use capabilities already provided by the host agent; specialized social, market, and data workflows can add SandBase providers when configured.

Quick Start

# Try it without installing: generate the complete Skill prompt
npx skills use sandbaseai/sandbase-skills@multi-source-search

# Or install it into Codex
npx skills add sandbaseai/sandbase-skills@multi-source-search --agent codex

# Use it with your agent's existing web/search tools
# "Fact-check this claim with independent sources and validate the evidence ledger"

multi-source-search needs no SandBase account when the host agent already provides search and page-reading tools. For specialized social, market, and data-provider Skills, set SANDBASE_API_KEY in your environment—never in a prompt or committed file.

GitHub CLI (official Agent Skills workflow)

GitHub CLI 2.90.0 or later can preview a Skill before installing it. Because this repository groups Skills by domain, pass the exact path shown in the catalog:

# Inspect the instructions and bundled files without installing anything
gh skill preview sandbaseai/sandbase-skills research/multi-source-search

# Install it for Codex at user scope
gh skill install sandbaseai/sandbase-skills research/multi-source-search \
  --agent codex --scope user

Replace the final path with another linked Skill path from the catalog below. See GitHub's gh skill documentation for supported agents, version pinning, updates, and security guidance.

Portable Agent Plugin (Copilot CLI and compatible clients)

Install the curated, account-free research plugin directly from this repository:

copilot plugin install sandbaseai/sandbase-skills:agent-plugin

The plugin follows the vendor-neutral Agent Plugins 1.0 specification and currently packages multi-source-search with its references and offline evidence-ledger validator. Its generated Skill copy is checked against the canonical source in CI.

Add SandBase MCP tools to the same agent

Skills provide repeatable instructions; SandBase CLI adds six MCP tools for discovering and running 2,000+ models and APIs. Connect one of the 25 supported client targets, then use these Skills and MCP tools in the same agent:

npx -y https://github.com/sandbaseai/cli/releases/download/v0.1.17/sandbaseai-cli-0.1.17.tgz connect --client codex

Replace codex with another client ID from the verified catalog.

DeepSeek Harness

Install all 88 Skills as a native DSH bundle:

dsh plugin --profile web add github:sandbaseai/sandbase-skills
dsh web

The bundle mounts the packaged marketing/ and research/ directories through DSH's filesystem Skill provider while preserving the default Skill roots.

To copy only the flagship Skill into one project's .dsh/skills/multi-source-search directory:

npx --yes github:sandbaseai/sandbase-skills add multi-source-search

The single-Skill installer and the bundle both run directly from the GitHub source, so no npm publication or SandBase account is required.

Claude Code marketplace

Install all 88 Skills as a native Claude Code plugin:

/plugin marketplace add sandbaseai/sandbase-skills
/plugin install sandbase-skills@sandbase-agent-skills

The marketplace manifest lists every Skill explicitly, so Claude Code can discover them on demand without copying directories by hand.

Verify research output offline

The multi-source-search Skill produces an evidence ledger that can be checked before you trust or share its synthesis:

See the complete worked example, which cross-checks one claim against primary documentation from GitHub, GitLab, and Atlassian.

python3 research/multi-source-search/scripts/validate_report.py \
  examples/verifiable-research-report.json
# VALID: 3 source(s), 1 claim(s), 2 provider(s)

The validator rejects unknown or duplicate sources, inflated confidence, unused evidence, and high-confidence claims that still have a declared conflict. It runs offline and checks internal consistency; it does not claim that a source is true.

For an end-to-end example that runs this research contract inside a governed, sandboxed runtime, read Build an Auditable Research Agent.

Try a Real Workflow

Install the matching Skill, then give your agent one of these tasks: