DEDUPED Hot AI repos, turned into use cases Get the Monday 5

How to check a research report's evidence ledger offline with Sandbase Skills

Run the bundled validator to catch unknown sources, inflated confidence, and unresolved conflicts.

Short answer: Clone the repository with git clone https://github.com/sandbaseai/sandbase-skills.git and open the folder; then open examples/branch-protection-research.md to see the worked example and what a checked claim looks like; run python3 research/multi-source-search/scripts/validate_report.py examples/verifiable-research-report.json. All steps.

  • Research
  • Intermediate
  • 30 minutes
  • From the README
  • Any agent

Authorization notice. Sandbase Skills is dual-use. Only use it on software, sites and systems you own or have explicit permission to test.

Who it is forA researcher or reviewer who has received an evidence-ledger report and must confirm it is internally consistent before relying on it.

The problemSynthesis from an agent can look rigorous while citing unused sources or overstating confidence.

The outcomeA pass or fail result from the validator that flags structural problems in the report.

What you need first

  • An AI agent that supports Agent Skills, such as Codex, Claude Code, Cursor, Gemini CLI, or DeepSeek Harness
  • Node.js with npx for the skills CLI installs
  • GitHub CLI 2.90.0 or later for the gh skill workflow
  • Python 3 and a local clone of the repository to run the offline validator
  • A SANDBASE_API_KEY set in your environment only for specialized social, market, and data-provider Skills

Steps

  1. Clone the repository with git clone https://github.com/sandbaseai/sandbase-skills.git and open the folder. (inferred, check the README)
  2. Open examples/branch-protection-research.md to see the worked example and what a checked claim looks like.
  3. Run python3 research/multi-source-search/scripts/validate_report.py examples/verifiable-research-report.json. (inferred, check the README)
  4. Compare your own report JSON against the same structure and fix any rejected sources or claims.

Prompt for your agent

In the sandbaseai/sandbase-skills repo, run research/multi-source-search/scripts/validate_report.py on <report.json>. Report the VALID or rejection messages, list each rejected source or claim, and verify that the validator checked internal consistency only, not whether any source is true.

Caveat. The validator checks internal consistency only and does not claim that any source is true, so you still need to read the sources.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for sandbaseai/sandbase-skills and "Check a research report's evidence ledger offline"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer