Cross-validate research with SandBase Multi-Source Search
Use SandBase Multi-Source Search to fact-check claims across independent web and academic sources, record disagreements, score confidence, and validate a structured evidence ledger offline.
npx skills add agentskillexchange/skills --skill cross-validate-research-with-sandbase-multi-source-search
# Cross-validate research with SandBase Multi-Source Search
What this skill actually does
Use the open-source SandBase `multi-source-search` Agent Skill when a decision or
fact-check needs more than one independent perspective. The workflow starts with
search and page-reading tools already available to the host agent, so a SandBase
account is not required. If SandBase tools are connected, it can add Tavily for
time-sensitive results, Exa for semantic discovery, Scholar for academic coverage,
and Cloudsway for broader web coverage. It treats retrieved pages as untrusted
evidence, traces derivative reporting back to common origins, keeps citations next
to claims, and explicitly records missing coverage and conflicting findings.
The Skill produces a JSON evidence ledger with source IDs, providers, claims,
supporting and contradicting evidence, confidence levels, and research gaps. Its
bundled offline validator checks URL shape, unique identifiers, provider diversity,
source references, and whether a confidence rating is supported by the declared
number of independent sources. This validates structural consistency rather than
asserting that a source is true. Use it for market research, technical comparisons,
current-event verification, or any report that must show how evidence supports each
conclusion.
## Installation
Install the maintained upstream Skill into Codex:
npm exec --package=skills@1.5.7 -- skills add sandbaseai/sandbase-skills --skill multi-source-search --agent codex
To inspect the workflow without installing it:
npm exec --package=skills@1.5.7 -- skills use sandbaseai/sandbase-skills@multi-source-search
The upstream repository also documents installation for Claude Code, Cursor,
Gemini CLI, and DeepSeek Harness. After completing research, validate the generated
ledger with the bundled script from the repository root:
python3 research/multi-source-search/scripts/validate_report.py research-report.json
## Source
– [SandBase Skills repository](https://github.com/sandbaseai/sandbase-skills)
– [Multi-Source Search Skill](https://github.com/sandbaseai/sandbase-skills/tree/main/research/multi-source-search)
– [Worked evidence-ledger example](https://github.com/sandbaseai/sandbase-skills/blob/main/examples/branch-protection-research.md)