← Back to Skills Marketplace
crftsmnd

Cross-Validate

by crftsmnd · GitHub ↗ · v1.1.0 · MIT-0
cross-platform ⚠ suspicious
121
Downloads
0
Stars
2
Active Installs
5
Versions
Install in OpenClaw
/install cross-validate
Description
Performs in-depth verification of Baseline-RAG claims using multiple sources and tests to provide a statistically tighter 95% confidence interval.
README (SKILL.md)

Cross-Validate

CI-Level 2 fact-checking with statistical confidence improvement.

What This Does

  • Takes results from baseline fact-checking
  • Performs deeper cross-validation using public sources
  • Uses more rigorous source verification
  • Returns confidence with tighter CI (90%+)

When to Use

Trigger AFTER Baseline-RAG runs, or explicitly with:

  • "cross-validate"
  • "verify with higher confidence"
  • "CI-95"

Workflow

Step 1: Receive Input

Get claim to verify from user or previous fact-check session.

Step 2: Public Sources

Use web_search and web_fetch (network tools) to find additional sources:

  • Academic papers (PubMed, Google Scholar)
  • Government/agency sources (.gov, .edu)
  • Recent news (last 6 months)
  • Peer-reviewed journals

Step 3: Cross-Reference

  • Check if additional sources agree/disagree
  • Weight by source credibility
  • Note publication dates

Step 4: Calculate Score

Heuristic formula (not statistically rigorous):

adjusted_score = min(95, base_score + (new_credible_sources × 5))

Note: This is a simple heuristic, not a confidence interval. Each additional credible source adds ~5 points, capped at 95 to leave room for uncertainty.

Step 5: Present Results

## Cross-Validation: [Claim]

### Scores
| Metric | Baseline | Cross-Validated |
|--------|---------|--------------|
| Score | [X]% | [Y]% |
| Sources | [N] | [M] |

### New Sources Added
- [source 1]
- [source 2]

### Verdict
[CONFIRMED / INCONCLUSIVE / REJECTED]

Tools Used

This skill uses platform tools:

  • web_search (for source discovery)
  • web_fetch (for source verification)

Not offline — requires network access for web searches.

Example

## Cross-Validation: "Coffee reduces cancer risk"

### Scores
| Metric | Baseline | Cross-Validated |
|--------|---------|--------------|
| Score | 65% | 85% |
| Sources | 3 | 8 |

### New Sources Added
- NIH.gov (2024)
- Google Scholar study
- WHO statement

### Verdict
CONFIRMED

Notes

  • Works fully offline with platform tools
  • No external dependencies
  • No payment required
  • Uses public sources only
Usage Guidance
Do not install or enable autonomous runs until the author clarifies the discrepancies. Specific questions to ask: (1) Is this skill free and platform-only as SKILL.md claims, or does it call an external paid API? (2) Which endpoint should the agent use (skill.yaml endpoint, README Netlify URL, or neither)? (3) Why do prices differ ($0.05 vs $0.30) and what is the required payment mechanism (x402-payment header appears in the function)? (4) Does the verification actually perform web_search/web_fetch, or does it return canned results from the external service? If you proceed, require explicit confirmation of payment behavior and audit network calls (block payments until verified). Prefer a version where the SKILL.md, skill.yaml, README, and code are reconciled and the external endpoint is a trusted, documented service before using with real data or payment credentials.
Capability Tags
cryptocan-make-purchases
Capability Assessment
Purpose & Capability
The skill description and SKILL.md state it runs using platform web_search/web_fetch and is free. But bundled artifacts (skill.yaml, README, and functions/verify.js) point to an external paid API/endpoint, list different prices ($0.05 in the function vs $0.30 in skill.yaml), and an external author_url. This mismatch between 'platform-only, no payment' and external paid endpoints is incoherent for the stated purpose.
Instruction Scope
SKILL.md instructs the agent to use platform tools (web_search, web_fetch) and does not mention external calls or payment. The repository contains a Netlify function that expects a payment header (x402-payment) and returns canned verification output instead of performing web searches. The runtime instructions therefore do not match the included code and hidden runtime behavior (payment gating, canned responses).
Install Mechanism
There is no install spec (instruction-only style), so nothing will be automatically written to disk or executed during install. The presence of code files is limited to a serverless function and config files; no installers or external archives are pulled in by the skill itself.
Credentials
The skill declares no required environment variables or credentials (which is consistent with SKILL.md). However, runtime artifacts indicate a paid external API and a custom payment header (x402-payment). While no secrets are requested, the skill's hidden requirement for a payment header / external endpoint is not documented in SKILL.md and is disproportionate to the 'free' claim.
Persistence & Privilege
The skill does not request persistent presence (always:false) and does not declare actions to modify other skills or system-wide settings. Autonomous invocation remains possible (platform default) but is not itself an additional red flag here.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install cross-validate
  3. After installation, invoke the skill by name or use /cross-validate
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.1.0
Updated price to sh.30, new endpoint
v1.0.3
- Clarified input: updated workflow to start with receiving a claim from user or previous session. - Added clearer explanation that the scoring formula is a simple heuristic, not a statistical confidence interval, and capped at 95%. - Updated workflow and tool descriptions to specify reliance on network access (no longer described as fully offline). - Improved tool usage explanation, clearly listing web_search and web_fetch and that no external services or API keys are needed.
v1.0.2
- Removed package.json and package-lock.json files. - No functional changes to core workflow or SKILL.md content. - Skill continues to require no external dependencies or services.
v1.0.1
Cross-validate 1.0.1 changelog: - Simplified workflow for cross-validation using only built-in web search and fetch tools. - Removed pricing and payment integration; now operates with no external dependencies or API keys. - Updated scoring logic for easier cross-validation, with an adjusted score formula and clearer thresholds. - Documentation streamlined: clarified trigger conditions, revised example outputs, and focused on source-based validation. - Cleaned up repository by removing obsolete deployment and server files.
v1.0.0
- Introduced "cross-validate" skill for advanced fact-checking with CI-95 confidence. - Deepens validation by cross-referencing multiple sources and applying statistical methods. - Provides significantly tighter confidence intervals compared to Baseline-RAG. - Includes clear workflow steps and result formatting with side-by-side comparison. - Outlines explicit triggers, usage instructions, and pricing notes for end users.
Metadata
Slug cross-validate
Version 1.1.0
License MIT-0
All-time Installs 2
Active Installs 2
Total Versions 5
Frequently Asked Questions

What is Cross-Validate?

Performs in-depth verification of Baseline-RAG claims using multiple sources and tests to provide a statistically tighter 95% confidence interval. It is an AI Agent Skill for Claude Code / OpenClaw, with 121 downloads so far.

How do I install Cross-Validate?

Run "/install cross-validate" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Cross-Validate free?

Yes, Cross-Validate is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Cross-Validate support?

Cross-Validate is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Cross-Validate?

It is built and maintained by crftsmnd (@crftsmnd); the current version is v1.1.0.

💬 Comments